Zhihao Jia

Kay Liu, Yingtong Dou, Yue Zhao et al.

Detecting which nodes in graphs are outliers is a relatively new machine learning task with numerous applications. Despite the proliferation of algorithms developed in recent years for this task, there has been no standard comprehensive setting for performance evaluation. Consequently, it has been difficult to understand which methods work well and when under a broad range of settings. To bridge this gap, we present--to the best of our knowledge--the first comprehensive benchmark for unsupervised outlier node detection on static attributed graphs called BOND, with the following highlights. (1) We benchmark the outlier detection performance of 14 methods ranging from classical matrix factorization to the latest graph neural networks. (2) Using nine real datasets, our benchmark assesses how the different detection methods respond to two major types of synthetic outliers and separately to "organic" (real non-synthetic) outliers. (3) Using an existing random graph generation technique, we produce a family of synthetically generated datasets of different graph sizes that enable us to compare the running time and memory usage of the different outlier detection algorithms. Based on our experimental results, we discuss the pros and cons of existing graph outlier detection algorithms, and we highlight opportunities for future research. Importantly, our code is freely available and meant to be easily extendable: https://github.com/pygod-team/pygod/tree/main/benchmark

Zhuoming Chen, Xinrui Zhong, Qilong Feng et al.

Sparse attention is becoming increasingly important for serving large language models (LLMs) as generation lengths continue to grow. However, deploying and evaluating new sparse attention algorithms at scale remains highly engineering-intensive, slowing both human researchers and AI agents in exploring the sparse attention design. To address this challenge, we present Vortex, a system that combines a Python-embedded frontend language atop a page-centric tensor abstraction for expressing a broad range of sparse attention algorithms, with an efficient backend tightly integrated into modern LLM serving stacks. Vortex enables rapid prototyping, deployment, and evaluation of sparse attention algorithms, effectively translating their theoretical efficiency gains into real-world throughput improvements. As a result, Vortex substantially accelerates the design and iteration of sparse attention algorithms. First, AI agents use Vortex to automatically generate and refine diverse algorithms, the best reaching up to $3.46\times$ higher throughput than full attention while preserving accuracy. Second, Vortex extends sparse attention to emerging architectures and very large models that are otherwise hard to experiment with, reaching up to $4.7\times$ higher throughput on the MLA-based GLM-4.7-Flash and $1.37\times$ on the 229B-parameter MiniMax-M2.7 on NVIDIA B200 GPUs.

Gabriele Oliaro, Yichao Fu, May Jiang et al.

LLM-based agents for GPU kernel generation are advancing rapidly, yet their progress is fundamentally constrained by the benchmarks they optimize against. Existing benchmarks are poorly aligned with production inference frameworks: they evaluate kernels on a single GPU with synthetic inputs, ignore the surrounding compilation stack, and reward replicating known optimizations rather than discovering new ones. The resulting reward signals are misleading: agents learn to generate kernels that score well in sandboxes but introduce interface incompatibilities, compilation-stack conflicts, and silent correctness degradation when integrated into real systems. We introduce FastKernels, a kernel benchmark built around a minimal set of 46 representative architectures spanning 8 categories, whose kernels collectively subsume those of 96.2% (409/425) of HuggingFace Transformers architectures. FastKernels doubles as a minimalistic, production-grade inference framework that runs at parity with hardened systems such as vLLM and SGLang on mainstream LLM serving and substantially exceeds upstream references on under-served architectures; each task's interface mirrors the corresponding module in the state-of-the-art library for its architecture family, enabling direct deployment of optimized kernels into production codebases. Evaluating state-of-the-art kernel agents on FastKernels, we find that even the strongest agent achieves only 0.94$\times$ aggregate speedup over production baselines, with weaker agents at $0.78\times$ and $0.53\times$ -- confirming that benchmark-production misalignment is a critical bottleneck for the field. We release FastKernels as a stepping stone toward kernel agents whose benchmark gains translate directly into production throughput improvements. Code is available at https://github.com/Snowflake-AI-Research/fastkernels

Yixuan Mei, Zikun Li, Zixuan Chen et al.

The usage of large language models (LLMs) has grown increasingly fragmented, with no single model dominating. Meanwhile, cloud providers offer a wide range of mid-tier and older-generation GPUs that enjoy better availability and deliver comparable performance per dollar to top-tier hardware. To efficiently harness these heterogeneous resources for serving multiple LLMs concurrently, we introduce Coral, an adaptive heterogeneity-aware multi-LLM serving system. The key idea behind Coral is to jointly optimize resource allocation and the serving strategy of each model replica across all models. To keep pace with shifting throughput demand and resource availability, Coral applies a lossless two-stage decomposition that preserves joint optimality while cutting online solve time from hours to tens of seconds. Our evaluation across 6 models and 20 GPU configurations shows that Coral reduces serving cost by up to 2.79$\times$ over the best baseline, and delivers up to 2.39$\times$ higher goodput under scarce resource availability.

Xupeng Miao, Chunan Shi, Jiangfei Duan et al.

The high computational and memory requirements of generative large language models (LLMs) make it challenging to serve them cheaply. This paper aims to reduce the monetary cost for serving LLMs by leveraging preemptible GPU instances on modern clouds, which offer accesses to spare GPUs at a much cheaper price than regular instances but may be preempted by the cloud at any time. Serving LLMs on preemptible instances requires addressing challenges induced by frequent instance preemptions and the necessity of migrating instances to handle these preemptions. This paper presents SpotServe, the first distributed LLM serving system on preemptible instances. Several key techniques in SpotServe realize fast and reliable serving of generative LLMs on cheap preemptible instances. First, SpotServe dynamically adapts the LLM parallelization configuration for dynamic instance availability and fluctuating workload, while balancing the trade-off among the overall throughput, inference latency and monetary costs. Second, to minimize the cost of migrating instances for dynamic reparallelization, the task of migrating instances is formulated as a bipartite graph matching problem, which uses the Kuhn-Munkres algorithm to identify an optimal migration plan that minimizes communications. Finally, to take advantage of the grace period offered by modern clouds, we introduce stateful inference recovery, a new inference mechanism that commits inference progress at a much finer granularity and allows SpotServe to cheaply resume inference upon preemption. We evaluate on real spot instance preemption traces and various popular LLMs and show that SpotServe can reduce the P99 tail latency by 2.4 - 9.1x compared with the best existing LLM serving systems. We also show that SpotServe can leverage the price advantage of preemptive instances, saving 54% monetary cost compared with only using on-demand instances.

John Thorpe, Pengzhan Zhao, Jonathan Eyolfson et al.

DNN models across many domains continue to grow in size, resulting in high resource requirements for effective training, and unpalatable (and often unaffordable) costs for organizations and research labs across scales. This paper aims to significantly reduce training costs with effective use of preemptible instances, i.e., those that can be obtained at a much cheaper price while idle, but may be preempted whenever requested by priority users. Doing so, however, requires new forms of resiliency and efficiency to cope with the possibility of frequent preemptions - a failure model that is drastically different from the occasional failures in normal cluster settings that existing checkpointing techniques target. We present Bamboo, a distributed system that tackles these challenges by introducing redundant computations into the training pipeline, i.e., whereby one node performs computations over not only its own layers but also over some layers in its neighbor. Our key insight is that training large models often requires pipeline parallelism where "pipeline bubbles" naturally exist. Bamboo carefully fills redundant computations into these bubbles, providing resilience at a low cost. Across a variety of widely used DNN models, Bamboo outperforms traditional checkpointing by 3.7x in training throughput, and reduces costs by 2.4x compared to a setting where on-demand instances are used.

Zikun Li, Jinjun Peng, Yixuan Mei et al.

Optimizing quantum circuits is challenging due to the very large search space of functionally equivalent circuits and the necessity of applying transformations that temporarily decrease performance to achieve a final performance improvement. This paper presents Quarl, a learning-based quantum circuit optimizer. Applying reinforcement learning (RL) to quantum circuit optimization raises two main challenges: the large and varying action space and the non-uniform state representation. Quarl addresses these issues with a novel neural architecture and RL-training procedure. Our neural architecture decomposes the action space into two parts and leverages graph neural networks in its state representation, both of which are guided by the intuition that optimization decisions can be mostly guided by local reasoning while allowing global circuit-wide reasoning. Our evaluation shows that Quarl significantly outperforms existing circuit optimizers on almost all benchmark circuits. Surprisingly, Quarl can learn to perform rotation merging, a complex, non-local circuit optimization implemented as a separate pass in existing optimizers.

Hongyi Jin, Bohan Hou, Guanjie Wang et al. · princeton

Modern GPU workloads, especially large language model (LLM) inference, suffer from kernel launch overheads and coarse synchronization that limit inter-kernel parallelism. Recent megakernel techniques fuse multiple operators into a single persistent kernel to eliminate launch gaps and expose inter-kernel parallelism, but struggle to handle dynamic shapes and data-dependent computation in real workloads. We present Event Tensor, a unified compiler abstraction for dynamic megakernels. Event Tensor encodes dependencies between tiled tasks, and enables first-class support for both shape and data-dependent dynamism. Built atop this abstraction, our Event Tensor Compiler (ETC) applies static and dynamic scheduling transformations to generate high-performance persistent kernels. Evaluations show that ETC achieves state-of-the-art LLM serving latency while significantly reducing system warmup overhead.

Liyan Zheng, Haojie Wang, Jidong Zhai et al.

Boosting the runtime performance of deep neural networks (DNNs) is critical due to their wide adoption in real-world tasks. Existing approaches to optimizing the tensor algebra expression of a DNN only consider expressions representable by a fixed set of predefined operators, missing possible optimization opportunities between general expressions. We propose OLLIE, the first derivation-based tensor program optimizer. OLLIE optimizes tensor programs by leveraging transformations between general tensor algebra expressions, enabling a significantly larger expression search space that includes those supported by prior work as special cases. OLLIE uses a hybrid derivation-based optimizer that effectively combines explorative and guided derivations to quickly discover highly optimized expressions. Evaluation on seven DNNs shows that OLLIE can outperform existing optimizers by up to 2.73$\times$ (1.46$\times$ on average) on an A100 GPU and up to 2.68$\times$ (1.51$\times$) on a V100 GPU, respectively.

Yue Huang, Chujie Gao, Siyuan Wu et al.

Generative Foundation Models (GenFMs) have emerged as transformative tools. However, their widespread adoption raises critical concerns regarding trustworthiness across dimensions. This paper presents a comprehensive framework to address these challenges through three key contributions. First, we systematically review global AI governance laws and policies from governments and regulatory bodies, as well as industry practices and standards. Based on this analysis, we propose a set of guiding principles for GenFMs, developed through extensive multidisciplinary collaboration that integrates technical, ethical, legal, and societal perspectives. Second, we introduce TrustGen, the first dynamic benchmarking platform designed to evaluate trustworthiness across multiple dimensions and model types, including text-to-image, large language, and vision-language models. TrustGen leverages modular components--metadata curation, test case generation, and contextual variation--to enable adaptive and iterative assessments, overcoming the limitations of static evaluation methods. Using TrustGen, we reveal significant progress in trustworthiness while identifying persistent challenges. Finally, we provide an in-depth discussion of the challenges and future directions for trustworthy GenFMs, which reveals the complex, evolving nature of trustworthiness, highlighting the nuanced trade-offs between utility and trustworthiness, and consideration for various downstream applications, identifying persistent challenges and providing a strategic roadmap for future research. This work establishes a holistic framework for advancing trustworthiness in GenAI, paving the way for safer and more responsible integration of GenFMs into critical applications. To facilitate advancement in the community, we release the toolkit for dynamic evaluation.

Zhihao Jia, Bing Wang, Changhao Chen

Neural rendering has garnered substantial attention owing to its capacity for creating realistic 3D scenes. However, its applicability to extensive scenes remains challenging, with limitations in effectiveness. In this work, we propose the Drone-NeRF framework to enhance the efficient reconstruction of unbounded large-scale scenes suited for drone oblique photography using Neural Radiance Fields (NeRF). Our approach involves dividing the scene into uniform sub-blocks based on camera position and depth visibility. Sub-scenes are trained in parallel using NeRF, then merged for a complete scene. We refine the model by optimizing camera poses and guiding NeRF with a uniform sampler. Integrating chosen samples enhances accuracy. A hash-coded fusion MLP accelerates density representation, yielding RGB and Depth outputs. Our framework accounts for sub-scene constraints, reduces parallel-training noise, handles shadow occlusion, and merges sub-regions for a polished rendering result. This Drone-NeRF framework demonstrates promising capabilities in addressing challenges related to scene complexity, rendering efficiency, and accuracy in drone-obtained imagery.

Mengdi Wu, Xiaoyu Jiang, Oded Padon et al.

This paper presents Prism, the first symbolic superoptimizer for tensor programs. The key idea is sGraph, a symbolic, hierarchical representation that compactly encodes large classes of tensor programs by symbolically representing some execution parameters. Prism organizes optimization as a two-level search: it constructs symbolic graphs that represent families of programs, and then instantiates them into concrete implementations. This formulation enables structured pruning of provably suboptimal regions of the search space using symbolic reasoning over operator semantics, algebraic identities, and hardware constraints. We develop techniques for efficient symbolic graph generation, equivalence verification via e-graph rewriting, and parameter instantiation through auto-tuning. Together, these components allow Prism to bridge the rigor of exhaustive search with the scalability required for modern ML workloads. Evaluation on five commonly used LLM workloads shows that Prism achieves up to $2.2\times$ speedup over best superoptimizers and $4.9\times$ over best compiler-based approaches, while reducing end-to-end optimization time by up to $3.4\times$.

Zhuoming Chen, Ranajoy Sadhukhan, Zihao Ye et al. · uw

Large language models (LLMs) with long context windows have gained significant attention. However, the KV cache, stored to avoid re-computation, becomes a bottleneck. Various dynamic sparse or TopK-based attention approximation methods have been proposed to leverage the common insight that attention is sparse. In this paper, we first show that TopK attention itself suffers from quality degradation in certain downstream tasks because attention is not always as sparse as expected. Rather than selecting the keys and values with the highest attention scores, sampling with theoretical guarantees can provide a better estimation for attention output. To make the sampling-based approximation practical in LLM generation, we propose MagicPIG, a heterogeneous system based on Locality Sensitive Hashing (LSH). MagicPIG significantly reduces the workload of attention computation while preserving high accuracy for diverse tasks. MagicPIG stores the LSH hash tables and runs the attention computation on the CPU, which allows it to serve longer contexts and larger batch sizes with high approximation accuracy. MagicPIG can improve decoding throughput by up to $5\times$ across various GPU hardware and achieve 54ms decoding latency on a single RTX 4090 for Llama-3.1-8B-Instruct model with a context of 96k tokens. The code is available at https://github.com/Infini-AI-Lab/MagicPIG.

Rui Pan, Yinwei Dai, Zhihao Zhang et al. · princeton

Recent advances in inference-time compute have significantly improved performance on complex tasks by generating long chains of thought (CoTs) using Large Reasoning Models (LRMs). However, this improved accuracy comes at the cost of high inference latency due to the length of generated reasoning sequences and the autoregressive nature of decoding. Our key insight in tackling these overheads is that LRM inference, and the reasoning that it embeds, is highly tolerant of approximations: complex tasks are typically broken down into simpler steps, each of which brings utility based on the semantic insight it provides for downstream steps rather than the exact tokens it generates. Accordingly, we introduce SpecReason, a system that automatically accelerates LRM inference by using a lightweight model to (speculatively) carry out simpler intermediate reasoning steps and reserving the costly base model only to assess (and potentially correct) the speculated outputs. Importantly, SpecReason's focus on exploiting the semantic flexibility of thinking tokens in preserving final-answer accuracy is complementary to prior speculation techniques, most notably speculative decoding, which demands token-level equivalence at each step. Across a variety of reasoning benchmarks, SpecReason achieves $1.4-3.0\times$ speedup over vanilla LRM inference while improving accuracy by $0.4-9.0\%$. Compared to speculative decoding without SpecReason, their combination yields an additional $8.8-58.0\%$ latency reduction. We open-source SpecReason at https://github.com/ruipeterpan/specreason.

Jingyuan Wang, Meiyan Xu, Zhihao Jia et al.

EEG foundation models (FMs) achieve strong cross-subject and cross-task generalization but impose substantial computational and memory costs that hinder deployment on embedded BCI systems. Knowledge distillation is a natural solution; however, conventional methods fail for EEG FMs because task-relevant semantics are often distributed across intermediate layers, and aggressive dimensionality reduction can distort oscillatory structure via representational collapse and aliasing. To address these challenges, we propose DLink (Distilling Layer-wise and Dominant Knowledge), a unified framework for transferring knowledge from large EEG FMs to compact students with three key innovations: (1) a dynamic Router that adaptively aggregates teacher layers to capture dominant intermediate representations; (2) an EEG MiC student with a Mimic-then-Compress pipeline, which inherits high-dimensional teacher features and then applies structured spatio-temporal compression to avoid a heavy classification head; and (3) spectral distillation that aligns teacher-student representations in the frequency domain to regularize compression and mitigate aliasing and temporal jitter. Experiments on four EEG benchmarks show that DLink enables compact students to outperform lightweight baselines while approaching fully fine-tuned FM performance at substantially lower model size and inference cost.

Ferdinand Kossmann, Zhihao Jia, Alex Aiken

The Mixture of Experts architecture allows for outrageously large neural networks by scaling model parameter size independently from computational demand (FLOPs). However, current DNN frameworks cannot effectively support the dynamic data flow in Mixture of Experts, and implementations on top of these frameworks need to use workarounds that introduce significant overheads. To address the limitation of these frameworks, we present DynaMoE, a DNN library that uses dynamic recompilations to optimize and adapt the use of computational resources to the dynamic needs of Mixture of Experts models. Our evaluation shows that DynaMoE achieves a 1.8x speedup and supports 2.3x larger model sizes when compared to existing MoE systems, even when not using recompilations. We then present further optimizations enabled by dynamic recompilations that yield an additional 1.7x speedup while simultaneously reducing memory pressure and improving model quality.

Zhihao Zhang, Zhuoming Chen, Heyang Huang et al.

As more practical and scalable quantum computers emerge, much attention has been focused on realizing quantum supremacy in machine learning. Existing quantum ML methods either (1) embed a classical model into a target Hamiltonian to enable quantum optimization or (2) represent a quantum model using variational quantum circuits and apply classical gradient-based optimization. The former method leverages the power of quantum optimization but only supports simple ML models, while the latter provides flexibility in model design but relies on gradient calculation, resulting in barren plateau (i.e., gradient vanishing) and frequent classical-quantum interactions. To address the limitations of existing quantum ML methods, we introduce Quark, a gradient-free quantum learning framework that optimizes quantum ML models using quantum optimization. Quark does not rely on gradient computation and therefore avoids barren plateau and frequent classical-quantum interactions. In addition, Quark can support more general ML models than prior quantum ML methods and achieves a dataset-size-independent optimization complexity. Theoretically, we prove that Quark can outperform classical gradient-based methods by reducing model query complexity for highly non-convex problems; empirically, evaluations on the Edge Detection and Tiny-MNIST tasks show that Quark can support complex ML models and significantly reduce the number of measurements needed for discovering near-optimal weights for these tasks.

Gabriele Oliaro, Zhihao Jia, Daniel Campos et al.

Speculative decoding is widely adopted to reduce latency in large language model (LLM) inference by leveraging smaller draft models capable of handling diverse user tasks. However, emerging AI applications, such as LLM-based agents, present unique workload characteristics: instead of diverse independent requests, agentic frameworks typically submit repetitive inference requests, such as multi-agent pipelines performing similar subtasks or self-refinement loops iteratively enhancing outputs. These workloads result in long and highly predictable sequences, which current speculative decoding methods do not effectively exploit. To address this gap, we introduce \emph{SuffixDecoding}, a novel method that utilizes efficient suffix trees to cache long token sequences from prompts and previous outputs. By adaptively speculating more tokens when acceptance likelihood is high and fewer when it is low, SuffixDecoding effectively exploits opportunities for longer speculations while conserving computation when those opportunities are limited. Evaluations on agentic benchmarks, including SWE-Bench and Text-to-SQL, demonstrate that SuffixDecoding achieves speedups of up to 5.3$\times$, outperforming state-of-the-art methods -- 2.8$\times$ faster than model-based approaches like EAGLE-2/3 and 1.9$\times$ faster than model-free approaches such as Token Recycling. SuffixDecoding is open-sourced at https://github.com/snowflakedb/ArcticInference

Xinle Cheng, Zhuoming Chen, Zhihao Jia

Diffusion models have revolutionized generative tasks, especially in the domain of text-to-image synthesis; however, their iterative denoising process demands substantial computational resources. In this paper, we present a novel acceleration strategy that integrates token-level pruning with caching techniques to tackle this computational challenge. By employing noise relative magnitude, we identify significant token changes across denoising iterations. Additionally, we enhance token selection by incorporating spatial clustering and ensuring distributional balance. Our experiments demonstrate reveal a 50%-60% reduction in computational costs while preserving the performance of the model, thereby markedly increasing the efficiency of diffusion models. The code is available at https://github.com/ada-cheng/CAT-Pruning

Zhihao Jia, Mingyi Jia, Junwen Duan et al.

Large Language Models (LLMs) demonstrate strong generalization and reasoning abilities, making them well-suited for complex decision-making tasks such as medical consultation (MC). However, existing LLM-based methods often fail to capture the dual nature of MC, which entails two distinct sub-tasks: symptom inquiry, a sequential decision-making process, and disease diagnosis, a classification problem. This mismatch often results in ineffective symptom inquiry and unreliable disease diagnosis. To address this, we propose \textbf{DDO}, a novel LLM-based framework that performs \textbf{D}ual-\textbf{D}ecision \textbf{O}ptimization by decoupling the two sub-tasks and optimizing them with distinct objectives through a collaborative multi-agent workflow. Experiments on three real-world MC datasets show that DDO consistently outperforms existing LLM-based approaches and achieves competitive performance with state-of-the-art generation-based methods, demonstrating its effectiveness in the MC task. The code is available at https://github.com/zh-jia/DDO.

Yixuan Mei, Yonghao Zhuang, Xupeng Miao et al.

This paper introduces Helix, a distributed system for high-throughput, low-latency large language model (LLM) serving in heterogeneous GPU clusters. The key idea behind Helix is to formulate inference computation of LLMs over heterogeneous GPUs and network connections as a max-flow problem on directed, weighted graphs, whose nodes represent GPU instances and edges capture both GPU and network heterogeneity through their capacities. Helix then uses a mixed integer linear programming (MILP) algorithm to discover highly optimized strategies to serve LLMs on heterogeneous GPUs. This approach allows Helix to jointly optimize model placement and request scheduling, two highly entangled tasks in heterogeneous LLM serving. Our evaluation on several heterogeneous clusters ranging from 24 to 42 GPU nodes shows that Helix improves serving throughput by up to 3.3x and reduces prompting and decoding latency by up to 66% and 24%, respectively, compared to existing approaches. Helix is available at https://github.com/Thesys-lab/Helix-ASPLOS25.

Mengdi Wu, Xinhao Cheng, Shengyu Liu et al.

We introduce Mirage, the first multi-level superoptimizer for tensor programs. A key idea in Mirage is $μ$Graphs, a uniform representation of tensor programs at the kernel, thread block, and thread levels of the GPU compute hierarchy. $μ$Graphs enable Mirage to discover novel optimizations that combine algebraic transformations, schedule transformations, and generation of new custom kernels. To navigate the large search space, Mirage introduces a pruning technique based on abstraction that significantly reduces the search space and provides a certain optimality guarantee. To ensure that the optimized $μ$Graph is equivalent to the input program, Mirage introduces a probabilistic equivalence verification procedure with strong theoretical guarantees. Our evaluation shows that Mirage outperforms existing approaches by up to 3.3$\times$ even for DNNs that are widely used and heavily optimized. Mirage is publicly available at https://github.com/mirage-project/mirage.

Xupeng Miao, Gabriele Oliaro, Zhihao Zhang et al.

This paper introduces SpecInfer, a system that accelerates generative large language model (LLM) serving with tree-based speculative inference and verification. The key idea behind SpecInfer is leveraging small speculative models to predict the LLM's outputs; the predictions are organized as a token tree, whose nodes each represent a candidate token sequence. The correctness of all candidate token sequences represented by a token tree is verified against the LLM in parallel using a novel tree-based parallel decoding mechanism. SpecInfer uses an LLM as a token tree verifier instead of an incremental decoder, which significantly reduces the end-to-end latency and computational requirement for serving generative LLMs while provably preserving model quality. Our evaluation shows that SpecInfer outperforms existing LLM serving systems by 1.5-2.8x for distributed LLM inference and by 2.6-3.5x for offloading-based LLM inference, while preserving the same generative performance. SpecInfer is publicly available at https://github.com/flexflow/FlexFlow/

Byungsoo Jeon, Sunghyun Park, Peiyuan Liao et al.

The strong demand for efficient and performant deployment of Deep Learning (DL) applications prompts the rapid development of a rich DL ecosystem. To keep up with this fast advancement, it is crucial for modern DL frameworks to efficiently integrate a variety of optimized tensor algebra libraries and runtimes as their backends and generate the fastest possible executable using these backends. However, current DL frameworks require significant manual effort and expertise to integrate every new backend while failing to unleash its full potential. Given the fast-evolving nature of the DL ecosystem, this manual approach often slows down continuous innovations across different layers; it prevents hardware vendors from the fast deployment of their cutting-edge libraries, DL framework developers must repeatedly adjust their hand-coded rules to accommodate new versions of libraries, and machine learning practitioners need to wait for the integration of new technologies and often encounter unsatisfactory performance. In this paper, we propose Collage, a DL framework that offers seamless integration of DL backends. Collage provides an expressive backend registration interface that allows users to precisely specify the capability of various backends. By leveraging the specifications of available backends, Collage automatically searches for an optimized backend placement strategy for a given workload and execution environment. Our evaluation shows that Collage outperforms the best existing framework for each hardware by $1.26\times$, $1.43\times$, $1.40\times$ on average on NVIDIA's RTX 2070 GPU, V100 GPU, and Intel's Xeon 8259CL CPU, respectively. Collage has been open-sourced and deployed in Apache TVM.

Yaoyao Ding, Ligeng Zhu, Zhihao Jia et al.

To accelerate CNN inference, existing deep learning frameworks focus on optimizing intra-operator parallelization. However, a single operator can no longer fully utilize the available parallelism given the rapid advances in high-performance hardware, resulting in a large gap between the peak performance and the real performance. This performance gap is more severe under smaller batch sizes. In this work, we extensively study the parallelism between operators and propose Inter-Operator Scheduler (IOS) to automatically schedule multiple operators' parallel execution through a novel dynamic programming algorithm. IOS consistently outperforms state-of-the-art libraries (e.g., TensorRT) by 1.1 to 1.5x on modern CNN benchmarks. The code to reproduce each experiment is available at: https://github.com/mit-han-lab/inter-operator-scheduler.

Zhuoming Chen, Avner May, Ruslan Svirschevski et al.

As the usage of large language models (LLMs) grows, performing efficient inference with these models becomes increasingly important. While speculative decoding has recently emerged as a promising direction for speeding up inference, existing methods are limited in their ability to scale to larger speculation budgets, and adapt to different hyperparameters and hardware. This paper introduces Sequoia, a scalable, robust, and hardware-aware algorithm for speculative decoding. To attain better scalability, Sequoia introduces a dynamic programming algorithm to find the optimal tree structure for the speculated tokens. To achieve robust speculative performance, Sequoia uses a novel sampling and verification method that outperforms prior work across different decoding temperatures. Finally, Sequoia introduces a hardware-aware tree optimizer that maximizes speculative performance by automatically selecting the token tree size and depth for a given hardware platform. Evaluation shows that Sequoia improves the decoding speed of Llama2-7B, Llama2-13B, and Vicuna-33B on an A100 by up to $4.04\times$, $3.73\times$, and $2.27\times$. For offloading setting on L40, Sequoia achieves as low as 0.56 s/token for exact Llama2-70B inference latency, which is $9.96\times$ on our optimized offloading system (5.6 s/token), $9.7\times$ than DeepSpeed-Zero-Inference, $19.5\times$ than Huggingface Accelerate.

Gabriele Oliaro, Xupeng Miao, Xinhao Cheng et al.

Finetuning large language models (LLMs) is essential for task adaptation, yet today's serving stacks isolate inference and finetuning on separate GPU clusters -- wasting resources and under-utilizing hardware. We introduce FlexLLM, the first system to co-serve LLM inference and PEFT-based finetuning on shared GPUs by fusing computation at the token level. FlexLLM's static compilation optimizations -- dependent parallelization and graph pruning significantly shrink activation memory, leading to end-to-end GPU memory savings by up to 80%. At runtime, a novel token-level finetuning mechanism paired with a hybrid token scheduler dynamically interleaves inference and training tokens within each co-serving iteration, meeting strict latency SLOs while maximizing utilization. In end-to-end benchmarks on LLaMA-3.1-8B, Qwen-2.5-14B, and Qwen-2.5-32B, FlexLLM maintains inference SLO compliance at up to 20 req/s, and improves finetuning throughput by $1.9-4.8\times$ under heavy inference workloads and $2.5-6.8\times$ under light loads, preserving over 76% of peak finetuning progress even at peak demand. FlexLLM is publicly available at https://flexllm.github.io.

Zhengxin Zhang, Dan Zhao, Xupeng Miao et al.

Finetuning large language models (LLMs) has been empirically effective on a variety of downstream tasks. Existing approaches to finetuning an LLM either focus on parameter-efficient finetuning, which only updates a small number of trainable parameters, or attempt to reduce the memory footprint during the training phase of the finetuning. Typically, the memory footprint during finetuning stems from three contributors: model weights, optimizer states, and intermediate activations. However, existing works still require considerable memory and none can simultaneously mitigate memory footprint for all three sources. In this paper, we present Quantized Side Tuing (QST), which enables memory-efficient and fast finetuning of LLMs by operating through a dual-stage process. First, QST quantizes an LLM's model weights into 4-bit to reduce the memory footprint of the LLM's original weights; QST also introduces a side network separated from the LLM, which utilizes the hidden states of the LLM to make task-specific predictions. Using a separate side network avoids performing backpropagation through the LLM, thus reducing the memory requirement of the intermediate activations. Furthermore, QST leverages several low-rank adaptors and gradient-free downsample modules to significantly reduce the trainable parameters, so as to save the memory footprint of the optimizer states. Experiments show that QST can reduce the total memory footprint by up to 2.3 $\times$ and speed up the finetuning process by up to 3 $\times$ while achieving competent performance compared with the state-of-the-art. When it comes to full finetuning, QST can reduce the total memory footprint up to 7 $\times$.

Zikun Li, Zhuofu Chen, Remi Delacourt et al.

Modern large language model (LLM) applications exhibit diverse service-level objectives (SLOs), from low-latency requirements in interactive coding assistants to more relaxed constraints in data wrangling tasks. Existing LLM serving systems, which rely on uniform batching and scheduling strategies, often fail to meet these heterogeneous SLOs concurrently. We present AdaServe, the first LLM serving system designed to support efficient multi-SLO serving through SLO-customized speculative decoding. AdaServe formulates multi-SLO serving as a constrained optimization problem and introduces a hardware-aware algorithm that constructs a speculation tree tailored to each request's latency target. It features a speculate-select-verify pipeline that enables fine-grained control over decoding speed while maximizing system throughput. AdaServe further adapts to workload variation by dynamically adjusting speculation parameters. Evaluations across diverse workloads show that AdaServe reduces SLO violations by up to 4.3$\times$ and improves goodput by up to 1.9$\times$ compared to the best performing baselines, highlighting its effectiveness in multi-SLO serving.

Xiaoxiang Shi, Colin Cai, Junjia Du et al.

Monolithic serving with chunked prefill improves GPU utilization by batching prefill and decode together, but suffers from fine-grained phase interference. Engine-level prefill-decode (PD) disaggregation avoids interference but incurs higher hardware and coordination overhead. Prior intra-GPU disaggregation approaches multiplex prefill and decode within a single GPU, using SLO-based tuning guided by heuristics from offline profiling or reactive feedback loops. However, these methods respond reactively to performance issues rather than anticipating them, limiting adaptability under dynamic workloads. We ask: can we achieve proactive intra-GPU disaggregation that adapts effectively to dynamic workloads? The key challenge lies in managing the conflicting resource demands of prefill and decode under varying conditions. We first show that GPU resources exhibit diminishing returns -- beyond a saturation point, more allocation yields minimal latency benefit. Second, we observe that memory bandwidth contention becomes a critical bottleneck. These insights motivate a design that dynamically partitions GPU resources across prefill and decode phases, while jointly considering compute capacity, memory footprint, and bandwidth contention. Evaluated on diverse LLMs and workloads, our system Nexus achieves up to 2.2x higher throughput, 20x lower TTFT, and 2.5x lower TBT than vLLM; outperforms SGLang by up to 2x; and matches or exceeds disaggregated vLLM.

Lijie Yang, Zhihao Zhang, Arti Jain et al.

Large reasoning models achieve strong performance through test-time scaling but incur substantial computational overhead, particularly from excessive token generation when processing short input prompts. While sparse attention mechanisms can reduce latency and memory usage, existing approaches suffer from significant accuracy degradation due to accumulated errors during long-generation reasoning. These methods generally require either high token retention rates or expensive retraining. We introduce LessIsMore, a training-free sparse attention mechanism for reasoning tasks, which leverages global attention patterns rather than relying on traditional head-specific local optimizations. LessIsMore aggregates token selections from local attention heads with recent contextual information, enabling unified cross-head token ranking for future decoding layers. This unified selection improves generalization and efficiency by avoiding the need to maintain separate token subsets per head. Evaluation across diverse reasoning tasks and benchmarks shows that LessIsMore preserves -- and in some cases improves -- accuracy while achieving a $1.1\times$ average decoding speed-up compared to full attention. Moreover, LessIsMore attends to $2\times$ fewer tokens without accuracy loss, achieving a $1.13\times$ end-to-end speed-up compared to existing sparse attention methods.

Jiawei He, Mingyi Jia, Zhihao Jia et al.

Retrieval-Augmented Large Language Models (LLMs), which integrate external knowledge, have shown remarkable performance in medical domains, including clinical diagnosis. However, existing RAG methods often struggle to tailor retrieval strategies to diagnostic difficulty and input sample informativeness. This limitation leads to excessive and often unnecessary retrieval, impairing computational efficiency and increasing the risk of introducing noise that can degrade diagnostic accuracy. To address this, we propose ICA-RAG (\textbf{I}nformation \textbf{C}ompleteness Guided \textbf{A}daptive \textbf{R}etrieval-\textbf{A}ugmented \textbf{G}eneration), a novel framework for enhancing RAG reliability in disease diagnosis. ICA-RAG utilizes an adaptive control module to assess the necessity of retrieval based on the input's information completeness. By optimizing retrieval and incorporating knowledge filtering, ICA-RAG better aligns retrieval operations with clinical requirements. Experiments on three Chinese electronic medical record datasets demonstrate that ICA-RAG significantly outperforms baseline methods, highlighting its effectiveness in clinical diagnosis.

Zhihao Jia, Qi Pang, Trung Tran et al.

In this work, we study the experts problem in the distributed setting where an expert's cost needs to be aggregated across multiple servers. Our study considers various communication models such as the message-passing model and the broadcast model, along with multiple aggregation functions, such as summing and taking the $\ell_p$ norm of an expert's cost across servers. We propose the first communication-efficient protocols that achieve near-optimal regret in these settings, even against a strong adversary who can choose the inputs adaptively. Additionally, we give a conditional lower bound showing that the communication of our protocols is nearly optimal. Finally, we implement our protocols and demonstrate empirical savings on the HPO-B benchmarks.

Byungsoo Jeon, Mengdi Wu, Shiyi Cao et al.

Deep neural networks (DNNs) continue to grow rapidly in size, making them infeasible to train on a single device. Pipeline parallelism is commonly used in existing DNN systems to support large-scale DNN training by partitioning a DNN into multiple stages, which concurrently perform DNN training for different micro-batches in a pipeline fashion. However, existing pipeline-parallel approaches only consider sequential pipeline stages and thus ignore the topology of a DNN, resulting in missed model-parallel opportunities. This paper presents graph pipeline parallelism (GPP), a new pipeline-parallel scheme that partitions a DNN into pipeline stages whose dependencies are identified by a directed acyclic graph. GPP generalizes existing sequential pipeline parallelism and preserves the inherent topology of a DNN to enable concurrent execution of computationally-independent operators, resulting in reduced memory requirement and improved GPU performance. In addition, we develop GraphPipe, a distributed system that exploits GPP strategies to enable performant and scalable DNN training. GraphPipe partitions a DNN into a graph of stages, optimizes micro-batch schedules for these stages, and parallelizes DNN training using the discovered GPP strategies. Evaluation on a variety of DNNs shows that GraphPipe outperforms existing pipeline-parallel systems such as PipeDream and Piper by up to 1.6X. GraphPipe also reduces the search time by 9-21X compared to PipeDream and Piper.

Ruslan Svirschevski, Avner May, Zhuoming Chen et al.

As large language models gain widespread adoption, running them efficiently becomes crucial. Recent works on LLM inference use speculative decoding to achieve extreme speedups. However, most of these works implicitly design their algorithms for high-end datacenter hardware. In this work, we ask the opposite question: how fast can we run LLMs on consumer machines? Consumer GPUs can no longer fit the largest available models (50B+ parameters) and must offload them to RAM or SSD. When running with offloaded parameters, the inference engine can process batches of hundreds or thousands of tokens at the same time as just one token, making it a natural fit for speculative decoding. We propose SpecExec (Speculative Execution), a simple parallel decoding method that can generate up to 20 tokens per target model iteration for popular LLM families. It utilizes the high spikiness of the token probabilities distribution in modern LLMs and a high degree of alignment between model output probabilities. SpecExec takes the most probable tokens continuation from the draft model to build a "cache" tree for the target model, which then gets validated in a single pass. Using SpecExec, we demonstrate inference of 50B+ parameter LLMs on consumer GPUs with RAM offloading at 4-6 tokens per second with 4-bit quantization or 2-3 tokens per second with 16-bit weights.

Zhihao Zhang, Alan Zhu, Lijie Yang et al.

Retrieval-augmented language models (RaLM) have demonstrated the potential to solve knowledge-intensive natural language processing (NLP) tasks by combining a non-parametric knowledge base with a parametric language model. Instead of fine-tuning a fully parametric model, RaLM excels at its low-cost adaptation to the latest data and better source attribution mechanisms. Among various RaLM approaches, iterative RaLM delivers a better generation quality due to a more frequent interaction between the retriever and the language model. Despite the benefits, iterative RaLM usually encounters high overheads due to the frequent retrieval step. To this end, we propose RaLMSpec, a speculation-inspired framework that provides generic speed-up over iterative RaLM while preserving the same model outputs through speculative retrieval and batched verification. By further incorporating prefetching, optimal speculation stride scheduler, and asynchronous verification, RaLMSpec can automatically exploit the acceleration potential to the fullest. For naive iterative RaLM serving, extensive evaluations over three language models on four downstream QA datasets demonstrate that RaLMSpec can achieve a speed-up ratio of 1.75-2.39x, 1.04-1.39x, and 1.31-1.77x when the retriever is an exact dense retriever, approximate dense retriever, and sparse retriever respectively compared with the baseline. For KNN-LM serving, RaLMSpec can achieve a speed-up ratio up to 7.59x and 2.45x when the retriever is an exact dense retriever and approximate dense retriever, respectively, compared with the baseline.

Xupeng Miao, Gabriele Oliaro, Zhihao Zhang et al.

In the rapidly evolving landscape of artificial intelligence (AI), generative large language models (LLMs) stand at the forefront, revolutionizing how we interact with our data. However, the computational intensity and memory consumption of deploying these models present substantial challenges in terms of serving efficiency, particularly in scenarios demanding low latency and high throughput. This survey addresses the imperative need for efficient LLM serving methodologies from a machine learning system (MLSys) research perspective, standing at the crux of advanced AI innovations and practical system optimizations. We provide in-depth analysis, covering a spectrum of solutions, ranging from cutting-edge algorithmic modifications to groundbreaking changes in system designs. The survey aims to provide a comprehensive understanding of the current state and future directions in efficient LLM serving, offering valuable insights for researchers and practitioners in overcoming the barriers of effective LLM deployment, thereby reshaping the future of AI.

Yue Zhao, George H. Chen, Zhihao Jia

Outlier detection (OD) is a key learning task for finding rare and deviant data samples, with many time-critical applications such as fraud detection and intrusion detection. In this work, we propose TOD, the first tensor-based system for efficient and scalable outlier detection on distributed multi-GPU machines. A key idea behind TOD is decomposing complex OD applications into a small collection of basic tensor algebra operators. This decomposition enables TOD to accelerate OD computations by leveraging recent advances in deep learning infrastructure in both hardware and software. Moreover, to deploy memory-intensive OD applications on modern GPUs with limited on-device memory, we introduce two key techniques. First, provable quantization speeds up OD computations and reduces its memory footprint by automatically performing specific floating-point operations in lower precision while provably guaranteeing no accuracy loss. Second, to exploit the aggregated compute resources and memory capacity of multiple GPUs, we introduce automatic batching, which decomposes OD computations into small batches for both sequential execution on a single GPU and parallel execution on multiple GPUs. TOD supports a diverse set of OD algorithms. Extensive evaluation on 11 real and 3 synthetic OD datasets shows that TOD is on average 10.9x faster than the leading CPU-based OD system PyOD (with a maximum speedup of 38.9x), and can handle much larger datasets than existing GPU-based OD systems. In addition, TOD allows easy integration of new OD operators, enabling fast prototyping of emerging and yet-to-discovered OD algorithms.

Zhihao Zhang, Zhihao Jia

A key challenge in neural architecture search (NAS) is quickly inferring the predictive performance of a broad spectrum of networks to discover statistically accurate and computationally efficient ones. We refer to this task as model performance inference (MPI). The current practice for efficient MPI is gradient-based methods that leverage the gradients of a network at initialization to infer its performance. However, existing gradient-based methods rely only on heuristic metrics and lack the necessary theoretical foundations to consolidate their designs. We propose GradSign, an accurate, simple, and flexible metric for model performance inference with theoretical insights. The key idea behind GradSign is a quantity Ψ to analyze the optimization landscape of different networks at the granularity of individual training samples. Theoretically, we show that both the network's training and true population losses are proportionally upper-bounded by Ψ under reasonable assumptions. In addition, we design GradSign, an accurate and simple approximation of Ψ using the gradients of a network evaluated at a random initialization state. Evaluation on seven NAS benchmarks across three training datasets shows that GradSign generalizes well to real-world networks and consistently outperforms state-of-the-art gradient-based methods for MPI evaluated by Spearman's ρ and Kendall's Tau. Additionally, we integrate GradSign into four existing NAS algorithms and show that the GradSign-assisted NAS algorithms outperform their vanilla counterparts by improving the accuracies of best-discovered networks by up to 0.3%, 1.1%, and 1.0% on three real-world tasks.

John Thorpe, Yifan Qiao, Jonathan Eyolfson et al.

A graph neural network (GNN) enables deep learning on structured graph data. There are two major GNN training obstacles: 1) it relies on high-end servers with many GPUs which are expensive to purchase and maintain, and 2) limited memory on GPUs cannot scale to today's billion-edge graphs. This paper presents Dorylus: a distributed system for training GNNs. Uniquely, Dorylus can take advantage of serverless computing to increase scalability at a low cost. The key insight guiding our design is computation separation. Computation separation makes it possible to construct a deep, bounded-asynchronous pipeline where graph and tensor parallel tasks can fully overlap, effectively hiding the network latency incurred by Lambdas. With the help of thousands of Lambda threads, Dorylus scales GNN training to billion-edge graphs. Currently, for large graphs, CPU servers offer the best performance-per-dollar over GPU servers. Just using Lambdas on top of CPU servers offers up to 2.75x more performance-per-dollar than training only with CPU servers. Concretely, Dorylus is 1.22x faster and 4.83x cheaper than GPU servers for massive sparse graphs. Dorylus is up to 3.8x faster and 10.7x cheaper compared to existing sampling-based systems.

Dheevatsa Mudigere, Yuchen Hao, Jianyu Huang et al.

Deep learning recommendation models (DLRMs) are used across many business-critical services at Facebook and are the single largest AI application in terms of infrastructure demand in its data-centers. In this paper we discuss the SW/HW co-designed solution for high-performance distributed training of large-scale DLRMs. We introduce a high-performance scalable software stack based on PyTorch and pair it with the new evolution of Zion platform, namely ZionEX. We demonstrate the capability to train very large DLRMs with up to 12 Trillion parameters and show that we can attain 40X speedup in terms of time to solution over previous systems. We achieve this by (i) designing the ZionEX platform with dedicated scale-out network, provisioned with high bandwidth, optimal topology and efficient transport (ii) implementing an optimized PyTorch-based training stack supporting both model and data parallelism (iii) developing sharding algorithms capable of hierarchical partitioning of the embedding tables along row, column dimensions and load balancing them across multiple workers; (iv) adding high-performance core operators while retaining flexibility to support optimizers with fully deterministic updates (v) leveraging reduced precision communications, multi-level memory hierarchy (HBM+DDR+SSD) and pipelining. Furthermore, we develop and briefly comment on distributed data ingestion and other supporting services that are required for the robust and efficient end-to-end training in production environments.

Zhihao Jia, Sina Lin, Rex Ying et al.

Graph Neural Networks (GNNs) are based on repeated aggregations of information across nodes' neighbors in a graph. However, because common neighbors are shared between different nodes, this leads to repeated and inefficient computations. We propose Hierarchically Aggregated computation Graphs (HAGs), a new GNN graph representation that explicitly avoids redundancy by managing intermediate aggregation results hierarchically, eliminating repeated computations and unnecessary data transfers in GNN training and inference. We introduce an accurate cost function to quantitatively evaluate the runtime performance of different HAGs and use a novel HAG search algorithm to find optimized HAGs. Experiments show that the HAG representation significantly outperforms the standard GNN graph representation by increasing the end-to-end training throughput by up to 2.8x and reducing the aggregations and data transfers in GNN training by up to 6.3x and 5.6x, while maintaining the original model accuracy.

Zhihao Jia, Sina Lin, Charles R. Qi et al.

The past few years have witnessed growth in the computational requirements for training deep convolutional neural networks. Current approaches parallelize training onto multiple devices by applying a single parallelization strategy (e.g., data or model parallelism) to all layers in a network. Although easy to reason about, these approaches result in suboptimal runtime performance in large-scale distributed training, since different layers in a network may prefer different parallelization strategies. In this paper, we propose layer-wise parallelism that allows each layer in a network to use an individual parallelization strategy. We jointly optimize how each layer is parallelized by solving a graph search problem. Our evaluation shows that layer-wise parallelism outperforms state-of-the-art approaches by increasing training throughput, reducing communication costs, achieving better scalability to multiple GPUs, while maintaining original network accuracy.