Serving Large Language Models (LLMs) is critical for AI-powered applications, yet it demands substantial computational resources, particularly in memory bandwidth and computational throughput. Low-precision computation has emerged as a key technique to improve efficiency while reducing resource consumption. Existing approaches for generating low-precision kernels are limited to weight bit widths that are powers of two and suffer from suboptimal performance because of high-level GPU programming abstractions. These abstractions restrict critical optimizations, such as fine-grained register management and optimized memory access patterns, that are essential for efficient low-precision computations. In this paper, we introduce Tilus, a domain-specific language designed for General-Purpose GPU (GPGPU) computing that supports low-precision data types with arbitrary bit widths from 1 to 8 while maintaining GPU programmability. Tilus features a thread-block-level programming model, a hierarchical memory space, a novel algebraic layout system, and extensive support for diverse low-precision data types. Tilus programs are compiled into highly efficient GPU programs through automatic vectorization and instruction selection. Extensive experiments demonstrate that Tilus efficiently supports a full spectrum of low-precision data types, and outperforms state-of-the-art low-precision kernels. Compared to existing compilers such as Triton and Ladder, as well as hand-optimized kernels such as QuantLLM and Marlin, Tilus achieves performance improvements of: 1.75x, 2.61x, 1.29x and 1.03x, respectively. We open-source Tilus at https://github.com/NVIDIA/tilus.

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.

Scaling modern deep learning workloads demands coordinated placement of data and compute across device meshes, memory hierarchies, and heterogeneous accelerators. We present Axe Layout, a hardware-aware abstraction that maps logical tensor coordinates to a multi-axis physical space via named axes. Axe unifies tiling, sharding, replication, and offsets across inter-device distribution and on-device layouts, enabling collective primitives to be expressed consistently from device meshes to threads. Building on Axe, we design a multi-granularity, distribution-aware DSL and compiler that composes thread-local control with collective operators in a single kernel. Experiments show that our unified approach can bring performance close to hand-tuned kernels on across latest GPU devices and multi-device environments and accelerator backends.

Recent large language model (LLM) agents have shown promise in using execution feedback for test-time adaptation. However, robust self-improvement remains far from solved: most approaches still treat each problem instance independently, without accumulating reusable knowledge. This limitation is particularly pronounced in domain-specific languages such as Triton, which are underrepresented in LLM pretraining data. Their strict constraints and non-linear optimization landscape further make naive generation and local refinement unreliable. We propose AdaExplore, an agent framework that enables self-improvement via accumulated execution feedback for performance-critical kernel code generation through two complementary stages: failure-driven adaptation and diversity-preserving search, jointly improving correctness and optimization performance without additional fine-tuning or external knowledge. In the adaptation stage, the agent synthesizes tasks and converts recurring failures into a reusable memory of validity rules, helping subsequent generations remain within the feasible set. In the search stage, the agent organizes candidate kernels as a tree and alternates between small local refinements and larger structural regeneration, allowing it to explore the optimization landscape beyond local optima. Experiments on kernel runtime optimization benchmarks validate these gains: AdaExplore achieves 3.12x and 1.72x speedups on KernelBench Level-2 and Level-3, respectively, within 100 steps, and continues to improve with additional computation.

Modern LLM agents increasingly rely on dynamic structured generation, such as tool calling and response protocols. Unlike traditional structured generation with static structures, these workloads vary both across requests and within a request, posing new challenges to existing engines. We present XGrammar-2, a structured generation engine for dynamic agentic workloads. Our design is based on two key ideas: first-class support for tag-triggered structure switching, and fine-grained reuse across requests with different output structures. Concretely, XGrammar-2 introduces TagDispatch for dynamic structural dispatching and Cross-Grammar Cache for substructure-level cache reuse across grammars. It further improves efficiency with an Earley-based adaptive token mask cache, just-in-time compilation, and repetition state compression. Experiments show that XGrammar-2 achieves over 6x faster compilation than prior structured generation engines, and incurs near-zero end-to-end overhead in modern LLM serving systems.

Agentic Variation Operators (AVO) are a new family of evolutionary variation operators that replace the fixed mutation, crossover, and hand-designed heuristics of classical evolutionary search with autonomous coding agents. Rather than confining a language model to candidate generation within a prescribed pipeline, AVO instantiates variation as a self-directed agent loop that can consult the current lineage, a domain-specific knowledge base, and execution feedback to propose, repair, critique, and verify implementation edits. We evaluate AVO on attention, among the most aggressively optimized kernel targets in AI, on NVIDIA Blackwell (B200) GPUs. Over 7 days of continuous autonomous evolution on multi-head attention, AVO discovers kernels that outperform cuDNN by up to 3.5

VIBETENSOR is an open-source research system software stack for deep learning, generated by LLM-powered coding agents under high-level human guidance. In this paper, "fully generated" refers to code provenance: implementation changes were produced and applied as agent-proposed diffs; validation relied on agent-run builds, tests, and differential checks, without per-change manual diff review. It implements a PyTorch-style eager tensor library with a C++20 core (CPU+CUDA), a torch-like Python overlay via nanobind, and an experimental Node.js/TypeScript interface. Unlike thin bindings, VIBETENSOR includes its own tensor/storage system, schema-lite dispatcher, reverse-mode autograd, CUDA runtime (streams/events/graphs), a stream-ordered caching allocator with diagnostics, and a stable C ABI for dynamically loaded operator plugins. We view this release as a milestone for AI-assisted software engineering: it shows coding agents can generate a coherent deep learning runtime spanning language bindings down to CUDA memory management, validated primarily by builds and tests. We describe the architecture, summarize the workflow used to produce and validate the system, and evaluate the artifact. We report repository scale and test-suite composition, and summarize reproducible microbenchmarks from an accompanying AI-generated kernel suite, including fused attention versus PyTorch SDPA/FlashAttention. We also report end-to-end training sanity checks on 3 small workloads (sequence reversal, ViT, miniGPT) on NVIDIA H100 (Hopper, SM90) and Blackwell-class GPUs; multi-GPU results are Blackwell-only and use an optional CUTLASS-based ring-allreduce plugin gated on CUDA 13+ and sm103a toolchain support. Finally, we discuss failure modes in generated system software, including a "Frankenstein" composition effect where locally correct subsystems interact to yield globally suboptimal performance.

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.

Mixture-of-Experts (MoE) has become the dominant architecture for frontier language models. To meet this demand, production frameworks have built optimized MoE training stacks over years of engineering effort. Yet evolving these stacks for new architectures and system optimizations remains expensive. With the rise of AI coding agents, they could automate parts of training-framework development and accelerate this evolution. But applying them to these existing frameworks carries hidden costs, invisible to today's throughput-only evaluations. We name this missing dimension agent-task efficiency (ATE): the cost of using coding agents to understand, operate, and extend a framework. Grounded in four agent-native design principles, we build PithTrain, a compact, agent-native MoE training framework. We further introduce ATE-Bench, covering real-world training-framework tasks. Our evaluation shows PithTrain matches the throughput of production frameworks, and on ATE-Bench, PithTrain enables higher agent-task efficiency, with up to 62

As agentic AI systems become increasingly capable of generating and optimizing GPU kernels, progress is constrained by benchmarks that reward speedup over software baselines rather than proximity to hardware-efficient execution. We present SOL-ExecBench, a benchmark of 235 CUDA kernel optimization problems extracted from 124 production and emerging AI models spanning language, diffusion, vision, audio, video, and hybrid architectures, targeting NVIDIA Blackwell GPUs. The benchmark covers forward and backward workloads across BF16, FP8, and NVFP4, including kernels whose best performance is expected to rely on Blackwell-specific capabilities. Unlike prior benchmarks that evaluate kernels primarily relative to software implementations, SOL-ExecBench measures performance against analytically derived Speed-of-Light (SOL) bounds computed by SOLAR, our pipeline for deriving hardware-grounded SOL bounds, yielding a fixed target for hardware-efficient optimization. We report a SOL Score that quantifies how much of the gap between a release-defined scoring baseline and the hardware SOL bound a candidate kernel closes. To support robust evaluation of agentic optimizers, we additionally provide a sandboxed harness with GPU clock locking, L2 cache clearing, isolated subprocess execution, and static analysis based checks against common reward-hacking strategies. SOL-ExecBench reframes GPU kernel benchmarking from beating a mutable software baseline to closing the remaining gap to hardware Speed-of-Light.

Recent advances show that large language models (LLMs) can act as autonomous agents capable of generating GPU kernels, but integrating these AI-generated kernels into real-world inference systems remains challenging. FlashInfer-Bench addresses this gap by establishing a standardized, closed-loop framework that connects kernel generation, benchmarking, and deployment. At its core, FlashInfer Trace provides a unified schema describing kernel definitions, workloads, implementations, and evaluations, enabling consistent communication between agents and systems. Built on real serving traces, FlashInfer-Bench includes a curated dataset, a robust correctness- and performance-aware benchmarking framework, a public leaderboard to track LLM agents' GPU programming capabilities, and a dynamic substitution mechanism (apply()) that seamlessly injects the best-performing kernels into production LLM engines such as SGLang and vLLM. Using FlashInfer-Bench, we further evaluate the performance and limitations of LLM agents, compare the trade-offs among different GPU programming languages, and provide insights for future agent design. FlashInfer-Bench thus establishes a practical, reproducible pathway for continuously improving AI-generated kernels and deploying them into large-scale LLM inference.

GPU kernel agents and GPU programming languages have advanced separately, leaving expert kernels difficult to reproduce. Agents usually treat the compiler as a fixed black box and receive only errors, correctness outcomes, and timing, while existing DSLs either hide critical scheduling decisions or expose them through difficult layout abstractions. We present CAKE, a compiler-agent co-design in which agents author CAKE IR, a typed, hardware-explicit schedule representation. CAKE exposes warp roles, memory movement, synchronization, and pipelines while supporting verification, cost modeling, and localized diagnostics. The harness itself evolves: recurring failures become verifier rules, IR primitives, model calibrations, and reusable optimization tactics. In matched implementation-hidden Flash-KMeans clean starts on B200, the best CAKE IR candidate at an 80-million-token budget runs at 1.144x the tuned FlashML baseline, compared with 0.928x for direct CUDA/PTX. Beyond this benchmark, agent-generated Kimi Delta Attention achieves a 2.05x geometric-mean speedup over official FlashKDA and passes end-to-end serving validation. Dispatcher-backed KNN and KMeans improve performance by 1.42x to 2.12x across more than 400 shapes, and four kernel changes are available as upstream PRs. CAKE targets NVIDIA GPUs from Ampere through Blackwell and separates single-shape evolution from library generalization and dispatch.

While existing machine learning (ML) frameworks focus on established platforms, like running CUDA on server-grade GPUs, there have been growing demands to enable emerging AI applications in a broader set of scenarios, such as running Large Language Models (LLMs) within browsers and mobile phones. However, deploying emerging models on new platforms (such as Metal and WebGPU) presents significant software engineering challenges due to rapid model evolution and limited tooling and practices for these platforms. Previous practice for ML model deployment often follows a bottom-up fashion, where engineers first implement individual required operators and then put them together. However, this traditional development approach fails to meet the productivity requirements when deploying emerging ML applications, with the testing and debugging part as a bottleneck. To this end, we introduce TapML, a top-down approach designed to streamline model deployment on diverse platforms. While the traditional bottom-up approach requires crafting manual tests, TapML automatically creates high-quality, realistic test data through operator-wise test carving. Furthermore, TapML uses a migration-based strategy to gradually offload model implementation from the mature source platform to the target platform, minimizing the debugging scope of compound errors. TapML has been used as the default development method in the MLC-LLM project to deploy emerging ML models. Within 2 years, TapML has accelerated the deployment of 105 emerging models in 27 model architectures across 5 emerging platforms. We show that TapML effectively boosts developer productivity while ensuring the quality of deployed models. Furthermore, we summarize comprehensive case studies from our real-world development, offering best practices for developing emerging ML systems.

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.

Transformers, driven by attention mechanisms, form the foundation of large language models (LLMs). As these models scale up, efficient GPU attention kernels become essential for high-throughput and low-latency inference. Diverse LLM applications demand flexible and high-performance attention solutions. We present FlashInfer: a customizable and efficient attention engine for LLM serving. FlashInfer tackles KV-cache storage heterogeneity using block-sparse format and composable formats to optimize memory access and reduce redundancy. It also offers a customizable attention template, enabling adaptation to various settings through Just-In-Time (JIT) compilation. Additionally, FlashInfer's load-balanced scheduling algorithm adjusts to dynamism of user requests while maintaining compatibility with CUDAGraph which requires static configuration. FlashInfer have been integrated into leading LLM serving frameworks like SGLang, vLLM and MLC-Engine. Comprehensive kernel-level and end-to-end evaluations demonstrate FlashInfer's ability to significantly boost kernel performance across diverse inference scenarios: compared to state-of-the-art LLM serving solutions, FlashInfer achieve 29-69 inter-token-latency reduction compared to compiler backends for LLM serving benchmark, 28-30 speedup for LLM serving with parallel generation.

Dynamic shape computations have become critical in modern machine learning workloads, especially in emerging large language models. The success of these models has driven demand for deploying them to a diverse set of backend environments. In this paper, we present Relax, a compiler abstraction for optimizing end-to-end dynamic machine learning workloads. Relax introduces first-class symbolic shape annotations to track dynamic shape computations globally across the program. It also introduces a cross-level abstraction that encapsulates computational graphs, loop-level tensor programs, and library calls in a single representation to enable cross-level optimizations. We build an end-to-end compilation framework using the proposed approach to optimize dynamic shape models. Experimental results on large language models show that Relax delivers performance competitive with state-of-the-art hand-optimized systems across platforms and enables deployment of emerging dynamic models to a broader set of environments, including mobile phones, embedded devices, and web browsers.

The applications of LLM Agents are becoming increasingly complex and diverse, leading to a high demand for structured outputs that can be parsed into code, structured function calls, and embodied agent commands. These developments bring significant demands for structured generation in LLM inference. Context-free grammar is a flexible approach to enable structured generation via constrained decoding. However, executing context-free grammar requires going through several stack states over all tokens in vocabulary during runtime, bringing non-negligible overhead for structured generation. In this paper, we propose XGrammar, a flexible and efficient structure generation engine for large language models. XGrammar accelerates context-free grammar execution by dividing the vocabulary into context-independent tokens that can be prechecked and context-dependent tokens that need to be interpreted during runtime. We further build transformations to expand the grammar context and reduce the number of context-independent tokens. Additionally, we build an efficient persistent stack to accelerate the context-dependent token checks. Finally, we co-design the grammar engine with LLM inference engine to overlap grammar computation with GPU executions. Evaluation results show that XGrammar can achieve up to 100x speedup over existing solutions. Combined with an LLM inference engine, it can generate near-zero overhead structure generation in end-to-end low-LLM serving.

Autoregressive Large Language Models (AR-LLMs) frequently exhibit implicit parallelism in sequential generation. Inspired by this, we introduce Multiverse, a new generative model enabling natively parallel generation. Multiverse internalizes a MapReduce paradigm, generating automatically through three stages: (i) a Map stage for adaptive task decomposition, (ii) a Process stage for parallel subtask execution, and (iii) a Reduce stage for lossless result synthesis. Next, we build a real-world Multiverse reasoning model with co-design of data, algorithm, and system, enabling rapid and seamless transfer from frontier AR-LLMs. For data creation, we develop Multiverse Curator, an automated LLM-assisted pipeline that transforms sequential reasoning chains into structured training data, avoiding costly human annotations. Algorithmically, we design Multiverse Attention to separate parallel reasoning steps while keeping compatibility with causal attention for efficient training. Systematically, we implement Multiverse Engine to support parallel inference. It features a dedicated interpreter that dynamically switches between sequential and parallel generation, triggered directly by the model. After a 3-hour fine-tuning with 1K examples, our Multiverse-32B stands as the only open-sourced non-AR model achieving performance on par with leading AR-LLMs of the same scale, evidenced by AIME24 & 25 scores of 54% and 46%, respectively. Moreover, our budget control experiments show that Multiverse-32B exhibits superior scaling, outperforming AR-LLMs by 1.87% on average using the same context length. Such scaling further leads to practical efficiency gain, achieving up to 2x speedup across varying batch sizes. We have open-sourced the entire Multiverse ecosystem, including data, model weights, serving system, supporting tools, as well as data curation prompts and detailed training and evaluation recipes.