🧩 并行策略
TP / EP / DCP 等张量、专家与上下文并行。
Finish RL Training Faster: Bitwise-Consistent, High-Performance, Agent-Driven
We make vLLM rollout and the training forward give bitwise identical logprobs on DeepSeek-V4-Flash and Nemotron-3.5-Lightning-30B-A3B (NVFP4), keep 91% of rollout throughput on DeepSeek-V4-Flash, and open-source the agent skill.
vLLM Support for NVIDIA Vera Rubin NVL72: 7.8x Throughput over GB200 NVL72
vLLM now runs on NVIDIA Vera Rubin NVL72 with day-0 model support, Rubin-tuned FlashInfer kernels, and locality-aware MoE, delivering 7.8x the per-GPU throughput of GB200 NVL72 on AgentX.
Announcing vllm-metal: Concurrent Serving on Apple Silicon
vllm-metal brings vLLM's paged, continuously batched serving stack to Apple Silicon, with flatter TTFT under concurrent agent load, batched MTP, and automatic M5 prefill acceleration.
vLLM x Novita AI: Chord, Faster INT4 MoE for Kimi K2.x. Up to 1.3x on H200, 2.15x on Untuned B300
Novita AI has open-sourced Chord, a high-performance W4A16 MoE CUDA kernel for Kimi K2.x serving shapes, with a Humming-compatible indexed path and grouped SM90 operators.
Kimi K3 Performance Optimizations in vLLM: The Road to 2.8× Throughput
Kimi K3 serving optimizations across scheduling, KDA prefix caching, ReplaySSM state recovery, PD disaggregation and state offload, parallelism, MoE, and GPU kernels.
Following the Bottleneck: Optimizing MiniMax M3 on AMD Instinct MI355X
A performance model for LLM serving: inspect local shapes, remove repeated work, verify data movement and dispatch, then follow the queue.
vLLM x AgentX: Optimizing for Real-World Agentic Serving
How vLLM optimizes KV cache management, parallelism, scheduling, and P/D disaggregation for agentic workloads, validated on SemiAnalysis AgentX with up to 130K tokens per GPU-second and a 14.6x-106x serving-cost advantage over Opus 5.
GLM 5.3 Optimizations, Part 1: Hybrid HiSparse Offloading in vLLM
vLLM integrates HiSparse as a pressure-driven memory tier that composes with the Hybrid Memory Allocator and KV offloading, letting GLM 5.3 requests keep decoding when their KV no longer fits in GPU memory, so concurrency stays high.
Serving LLMs on Tenstorrent Hardware: Inside the vLLM TT Plugin
Tenstorrent accelerators join vLLM as an out-of-tree platform plugin, driven by mesh-architecture choices: phase-based scheduling, single-process data parallelism on Galaxy, on-device sampling with host fallback, and async decode overlap.
MiniMax H3 on vLLM-Omni: From System-Wide Optimization to Real-Time Serving with FastVideo’s FastH3
How vLLM-Omni optimizes and scales the complete MiniMax H3 stack, then integrates FastVideo’s four-step FastH3 for generation faster than playback.
Large-Scale Sharded Weight Transfer with Ray Direct Transport (RDT) in vLLM
We implement a native sharded weight transfer engine in vLLM utilizing Ray Direct Transport (RDT), achieving weight transfer for the Kimi K2 model in BF16 on 48 8xH100 nodes in 7.53s
IsoExec: Unified Execution to Eliminate Trainer-Inference Mismatch in SkyRL
IsoExec unifies numerical execution across SkyRL's vLLM and Megatron runtimes, reducing the average rollout-versus-training logprob difference below 1e-6 on Qwen3.5-35B-A3B with 25% overhead.
Distributed Layerwise Offload: Scaling Toward 200B+ DiT Models Efficiently in vLLM-Omni
Distributed Layerwise Offload shards and streams DiT weights across devices, serving a measured 124 GB Cosmos3 model on 64 GB HBM and estimating a path toward 200B+ models.
Adaptive Verification in vLLM: DSpark confidence-scheduled verification
本文是『推测解码』路径中自适应验证主题的第一篇，讲 DSpark 如何用置信度头按存活概率调度验证预算；下一篇是无 confidence head 的在线估计器方案，建议对比阅读以理解两类自适应验证的取舍。
Day 0 Support for Qwen3.8-2.4T-A95B on vLLM
Day-0 vLLM support for Qwen3.8-2.4T-A95B: a 2.4-trillion-parameter hybrid MoE model served out of the box, with FP8/BF16 checkpoints plus NVFP4 and MXFP4 quantized weights, and co-developed kernels on NVIDIA and AMD hardware.
Efficient Decode Context Parallelism with vLLM for Long Context Workloads
Decode Context Parallelism (DCP) in vLLM shards KV cache across GPUs by sequence dimension, enabling 3× higher throughput on long-context agentic workloads compared to standard tensor parallelism.
vLLM Reaches 25K Total TPS/GPU on Qwen3.5
本文是『PD 分离服务』路径末尾三篇生产规模实战中建议最先读的一篇：它是 Qwen3.8 篇的直接前置（后者开篇引用本文结果），也是 GB200/GB300 NVL72 + 混合 GDN 架构 PD serving 的基础认知入口，可再顺读 GLM-5.2 篇对照 SLA 导向的取舍。
Kimi K3 Is Here: Efficient Day-0 Support on vLLM
vLLM delivers day-0 Kimi K3 serving with hybrid KDA prefix caching, DSpark speculative decoding, production-scale disaggregation, and optimized kernels across NVIDIA and AMD GPUs.
From Day 0 to Production SLAs: Serving GLM-5.2 on 24 NVIDIA B300 GPUs with vLLM
本文是『PD 分离服务』学习路径末尾三篇生产规模实战之一，以 SLA 合规而非峰值吞吐为第一目标，与两篇 Qwen 吞吐导向文章形成方法论对照。建议在读完 Qwen3.5（25K TPS）篇建立 NVL72 + 混合架构 PD 基础后阅读，重点体会约束驱动的配置取舍与边界问题定位。
A Preview of Production-Scale Kimi K3 Support on vLLM
A preview of production-scale Kimi K3 support in vLLM, including KDA-aware prefix caching, fused kernels, optimized MXFP4 MoE, multimodal integration, and initial NVIDIA and AMD paths.
TML Inkling on vLLM: Day-0 Support with Optimized Performance
vLLM brings day-0 support to TML Inkling, a 1T-parameter multimodal model, with MTP, long-context serving, parallelism, and up to 380 tokens per second per user on NVIDIA GB200 GPUs.
vime + ROCm: End-to-End RL Post-Training on AMD Instinct™ GPUs
Announcing ROCm support for vime, now running end-to-end on AMD Instinct MI355X GPUs with prebuilt container.
MiniMax M3 in vLLM: Day-0 Serving for 1M-Token Multimodal Reasoning
How vLLM serves MiniMax M3 with MiniMax Sparse Attention, multimodal and reasoning parsers, MXFP8 weights, and long-context deployment recipes.
Announcing vime: A Simple, Stable, and Efficient RL Framework for LLMs
vime connects slime's training stack with vLLM rollouts to provide a simple, stable, and efficient RL post-training pipeline.
Accelerating vLLM-Omni Inference with AutoRound Quantization
How AutoRound integrates with vLLM-Omni to serve W4A16 quantized multimodal, diffusion, image, and video models with smaller checkpoints, preserved quality, Intel XPU acceleration, and NVIDIA GPU support.
vLLM x Novita AI: PegaFlow for Production-Grade External KV Cache
How PegaFlow integrates with vLLM as an external KV cache service, using a Rust daemon, CUDA IPC, RDMA, SSD caching, and the external KV connector to improve startup, sharing, throughput, and cache lifecycle.
Elastic Expert Parallelism in vLLM
How Elastic Expert Parallelism lets vLLM scale Mixture-of-Experts serving up or down at runtime by changing data-parallel workers, redistributing experts, and coordinating live topology changes without server restarts.
vLLM Tops the Artificial Analysis Leaderboard
How vLLM achieved leading Artificial Analysis results for DeepSeek V3.2, MiniMax-M2.5, and Qwen 3.5 397B using open-source kernel fusion, speculative decoding, Blackwell optimizations, and model-specific serving work.
Serving Agentic Workloads at Scale with vLLM x Mooncake
How vLLM integrates Mooncake Store as a distributed KV cache for agentic workloads, reusing shared prefixes across turns and instances to improve throughput, TTFT, end-to-end latency, and multi-GPU scaling.
Disaggregated Serving for Hybrid SSM Models in vLLM
『PD 分离服务』路径的进阶深化篇:建议先读 vLLM 基础 PD 分离文章(介绍 NIXL connector 的 v0-disagg-prefill)掌握标准流程,再读本文理解 hybrid SSM 模型带来的底层改造;可与 AFD Plugin(按组件拆分)、TileRT(decode 可插拔)两篇生态文对照阅读。
Next-Level Inference: Why Your Single-Node vLLM Setup Needs Prefill-Decode Disaggregation
How single-node prefill/decode disaggregation in vLLM uses AMD MORI-IO on an 8-GPU MI300X node to separate prefill and decode, transfer KV cache efficiently, stabilize ITL, and improve goodput.
DeepSeek-V3.2 on GB300: Performance Breakthrough
What DeepSeek-V3.2 and DeepSeek-R1 benchmark results show on NVIDIA GB300 with vLLM, covering NVFP4 quantization, TP and EP deployment, throughput, and reproducible setup details.
Driving vLLM WideEP and Large-Scale Serving Toward Maturity on Blackwell (Part I)
How vLLM improves WideEP and large-scale DeepSeek-style MoE serving on NVIDIA GB200 with NVFP4 and FP8 kernels, fusion, prefill/decode disaggregation, weight offloading, and reduced chunking overhead.
GPT-OSS Performance Optimizations on NVIDIA Blackwell: Pushing the Pareto Frontier
How vLLM and NVIDIA optimized GPT-OSS on Blackwell with FlashInfer, torch.compile fusion, FP8 KV cache, async scheduling, stream interval tuning, and deployment recipes that improve throughput and interactivity.
vLLM Large Scale Serving: DeepSeek @ 2.2k tok/s/H200 with Wide-EP
How vLLM reaches 2.2k tokens per second per H200 for DeepSeek-style MoE serving with Wide-EP, async scheduling, dual-batch overlap, disaggregated serving, CUDA graphs, DeepGEMM, and expert load balancing.
Encoder Disaggregation for Scalable Multimodal Model Serving
How vLLM EPD separates visual encoders from text prefill and decode, covering LMM serving, GPU resource scaling, multimodal interference, and pipelined execution.
vLLM Router: A High-Performance and Prefill/Decode Aware Load Balancer for Large-scale Serving
『PD 分离服务』学习路径的入门/生态篇:篇幅短,快速建立『P/D 分离在生产中需要专门编排层』的认知;建议先于深度实操文章(如 disaggregated serving guide)阅读,为理解 router/proxy 在 P/D 架构中的角色打基础。
Diving into speculative decoding training support for vLLM with Speculators v0.3.0
『推测解码』路径的第二篇（工程进阶）：承接第一篇对 EAGLE/Medusa 的概念铺垫，深入 Eagle3 draft model 的训练数据构造、训练技术与 vLLM 部署细节，建议在读过入门篇后、读 EAGLE 3.1 之前阅读。
Shared Memory IPC Caching: Accelerating Data Transfer in LLM Inference Systems
How shared memory IPC caching in vLLM reduces redundant data transfers for multimodal and multi-process inference, improving prefill throughput and TTFT by sharing large inputs across coordinator and worker processes.
Fast and Affordable LLMs serving on Intel Arc Pro B-Series GPUs with vLLM
How vLLM serves LLMs on Intel Arc Pro B-Series GPUs with MoE optimizations, persistent kernels, multi-GPU scaling, LoRA, speculative decoding, structured outputs, and mixed-precision recipes.
From Monolithic to Modular: Scaling Semantic Routing with Extensible LoRA
How vLLM Semantic Router refactors its Rust classification layer with modular model support, Qwen3-Embedding, EmbeddingGemma, LoRA-based multi-task classification, and concurrent routing execution.
Zero-Reload Model Switching with vLLM Sleep Mode
How vLLM Sleep Mode enables fast model switching by hibernating weights to CPU RAM or discarding them while preserving process state, CUDA graphs, allocators, and kernel warmup to avoid full reloads.
vLLM TPU: A New Unified Backend Supporting PyTorch and JAX on TPU
How the redesigned vLLM TPU backend uses tpu-inference, JAX-to-XLA lowering, Torchax, ragged paged attention, and unified PyTorch and JAX support to improve TPU performance and model coverage.
SemiAnalysis InferenceMAX: vLLM and NVIDIA Accelerate Blackwell Inference
How vLLM and NVIDIA optimize Blackwell inference for SemiAnalysis InferenceMAX, improving gpt-oss 120B and Llama 3.3 70B throughput with FP4 kernels, scheduling work, and Pareto-frontier benchmarking.
DeepSeek-V3.2-Exp in vLLM: Fine-Grained Sparse Attention in Action
How vLLM supports DeepSeek-V3.2-Exp with DeepSeek Sparse Attention, lightning indexer caches, separate prefill and decode paths, FlashMLA sparse attention, DeepGEMM kernels, and Blackwell deployment.
Inside vLLM: Anatomy of a High-Throughput LLM Inference System
How vLLM's inference engine works, covering PagedAttention, continuous batching, prefix caching, speculative decoding, multi-GPU serving, scheduling, and benchmarking for high-throughput LLM workloads.
Introduction to torch.compile and How It Works with vLLM
How torch.compile works inside vLLM, including TorchDynamo graph capture, TorchInductor code generation, custom compiler passes, graph breaks, model compilation strategy, and performance optimization.
vLLM Now Supports gpt-oss
How vLLM supports gpt-oss 20B and 120B on NVIDIA Blackwell, Hopper, and AMD GPUs, with MXFP4 MoE kernels, efficient attention, hybrid KV cache allocation, and built-in tool support.
Accelerating RLHF with vLLM, Best Practice from OpenRLHF
How OpenRLHF uses vLLM, Ray, ZeRO-3, AutoTP, Ray placement groups, and weight synchronization to accelerate PPO and RLHF sample generation for reasoning models with long chain-of-thought outputs.
Llama 4 in vLLM
How vLLM serves Meta Llama 4 Scout and Maverick multimodal MoE models with long-context support, tensor parallel deployment, H100 and H200 guidance, FP8 variants, and performance tips.
PTPC-FP8: Boosting vLLM Performance on AMD ROCm
How PTPC-FP8 quantization improves vLLM performance on AMD ROCm by combining per-token activation scaling and per-channel weight scaling for near-BF16 accuracy with FP8 speed.
Distributed Inference with vLLM
A guide to distributed inference in vLLM, covering tensor parallelism, pipeline parallelism, multi-GPU and multi-node serving, KV cache challenges, speculative decoding, communication kernels, and control-plane design.
vLLM V1: A Major Upgrade to vLLM's Core Architecture
What changed in vLLM V1: a re-architected engine with a simpler scheduler, near-zero-overhead prefix caching, cleaner tensor parallelism, multiprocessing API server, and default optimizations for higher-throughput serving.
vLLM 2024 Retrospective and 2025 Vision
A vLLM 2024 retrospective and 2025 roadmap covering community growth, model and hardware support, production adoption, office hours, ecosystem partnerships, and the path toward universal open-source serving.
Serving LLMs on AMD MI300X: Best Practices
Best practices for serving LLMs with vLLM on AMD MI300X, covering ROCm setup, Llama 3.1 70B and 405B benchmarks, chunked prefill, multi-step scheduling, prefix caching, graph capture, and AMD tuning.
How Speculative Decoding Boosts vLLM Performance by up to 2.8x
『推测解码』学习路径的第一篇（入门）：先读它建立 draft/target 协作、n-gram、Medusa/EAGLE 等基础概念和 vLLM 系统架构认知，再进入后两篇分别学习 Eagle3 draft model 的训练工程与 EAGLE 3.1 的算法改进。
vLLM’s Open Governance and Performance Roadmap
vLLM's open governance and performance roadmap, covering LF AI and Data incubation, public benchmarks, optimized kernels, async scheduling, API frontend overhead, torch.compile, disaggregated prefill, and community research.
Announcing Llama 3.1 Support in vLLM
How vLLM supports Meta Llama 3.1 models, including 128K context, Llama 3.1 405B serving, chunked prefill, FP8 quantization, tensor and pipeline parallelism, CPU offloading, and early performance results.