T
traeai
Sign in

产品

vLLM

别名:vlm

开源推理引擎,支持多芯片架构

已跟踪 30 条高相关材料

TraeAI 观察

相关材料

已收录 30 条与 vLLM 相关的内容,按评分排序。

How Trustpilot built a real-time architecture for data enrichment using Gemma

How Trustpilot built a real-time architecture for data enrichment using Gemma

Google Cloud Blog992 字 (约 4 分钟)
92

Trustpilot built a real-time data enrichment pipeline using fine-tuned Gemma models to process millions of reviews under strict latency and cost constraints, achieving near-teacher-model accuracy with full control.

入选理由:采用 google/gemma-2-9b 基础模型,通过共识标注生成高质量训练集,微调后准确率仅比教师模型低几个百分点。

FeaturedArticle#Gemma#Dataflow#LLM#Real-time Architecture#Fine-tuning英文
英伟达重新思考AI TCO:为何每Token成本才是唯一重要的指标

NVIDIA advocates for cost per token as the core economic metric for AI infrastructure, replacing traditional measures like compute cost or FLOPS per dollar, emphasizing full-stack optimization to reduce inference costs and enhance business value.

入选理由:每Token成本是衡量AI基础设施经济效益的核心指标,直接反映实际产出效率。

FeaturedArticle#NVIDIA#AI TCO#Inference Optimization#Cost Per Token中文
Build real-time voice applications with Amazon SageMaker AI and vLLM

Build Real-Time Voice Applications with Amazon SageMaker AI and vLLM

AWS Machine Learning Blog2911 字 (约 12 分钟)
87

AWS combines SageMaker AI with vLLM to enable bidirectional streaming speech-to-text inference, supporting real-time voice assistants, live captions, and more with significantly reduced latency.

入选理由:SageMaker AI提供原生HTTP/2双向流式传输(端口8443),自动处理HTTP/2事件流与WebSocket协议转换

FeaturedArticle#AWS#SageMaker#vLLM#Voice AI#Streaming Inference英文
How Open Source Became AI's Backbone | Inferact with a16z

How Open Source Became AI's Backbone | Inferact with a16z

a16z11384 字 (约 46 分钟)
85

开源推理引擎vLLM通过10倍性能优化和跨芯片兼容性,正在重塑AI基础设施,可能缩小与闭源模型的差距。

入选理由:vLLM支持Nvidia/AMD/Google最新芯片,实现跨平台部署

FeaturedVideo#vLLM#开源AI#推理引擎#a16z中英混合
Together AI Blog 图标

ThunderAgent通过优化KV缓存管理,实现单节点吞吐量提升2倍,集群加速2.4倍,解决代理推理中的缓存抖动问题。

入选理由:单节点吞吐量提升2.5倍,P50延迟降低10倍

FeaturedArticle#ThunderAgent#合成数据生成#LLM推理优化#KV缓存管理英文
148. 对游凯超3小时访谈:开源Infra、和模型Co-design 、“如果vLLM失败,我们会后悔一辈子”

开源项目vLLM的商业化与模型基础设施协同设计为AI领域提供了新思路,揭示了开源社区向企业转型的挑战与技术融合路径。

入选理由:vLLM从论文到开源社区耗时3年,社区维护者放弃个人利益坚持开源精神。

FeaturedPodcast#开源#AI Infra#模型协同设计#创业公司#vLLM中文
Google Developers Blog 图标

Run Ray on TPU, Part 2: Ray AI libraries

Google Developers Blog1659 字 (约 7 分钟)
85

在TPU上部署Ray AI库需通过topology字段确保多主机模型正确分配资源,避免因跨slice通信失败导致部署停滞。

入选理由:设置topology字段可防止多主机模型跨slice部署,避免集体通信失败

FeaturedArticle#TPU#Ray#GKE#AI库#vLLM英文
Driving the Future of Open Source AI: An Update from PyTorch Foundation Projects

PyTorch基金会宣布成立多项目基金会,PyTorch 2.13发布显著优化性能,vLLM推出Model Runner V2并公布2026路线图。

入选理由:PyTorch 2.13在Apple Silicon上实现FlexAttention性能提升至SDPA的12倍

FeaturedArticle#PyTorch#开源AI#模型优化#DeepSpeed#vLLM英文
Open Models are ready for agents. Their APIs are not.

Open Models are ready for agents. Their APIs are not.

Mozilla AI Blog969 字 (约 4 分钟)
85

开源模型已具备代理应用能力,但API兼容性不足成为生产环境瓶颈,Mozilla提出开源网关Otari解决该问题。

入选理由:开源模型推理能力已满足代理产品需求,但API兼容性仅支持基础聊天功能

FeaturedArticle#AI代理#开源模型#API兼容性#Otari英文
KDnuggets 图标

5 Agentic Workflows to Automate Your Data Science Pipeline

KDnuggets5486 字 (约 22 分钟)
85

自动化数据科学流程的5种代理工作流可节省45%时间,Databricks已集成相关功能,核心依赖ReAct循环与LLM工具。

入选理由:数据科学家45%时间消耗在数据清洗,代理可自动化处理

FeaturedArticle#AI#数据科学#自动化#MLOps#Python英文
Hugging Face Blog 图标

Run a vLLM Server on HF Jobs in One Command

Hugging Face Blog1704 字 (约 7 分钟)
85

Hugging Face 提供了一种快速部署 OpenAI 兼容 LLM 的方法,仅需一条命令即可完成。

入选理由:使用 hf jobs run 命令可在 Hugging Face 上一键部署 LLM 服务。

FeaturedArticle#Hugging Face#vLLM#LLM#部署英文
Query Your Codebase with DeepSeek V4 and vLLM

Query Your Codebase with DeepSeek V4 and vLLM

NVIDIA Developer539 字 (约 3 分钟)
85

DeepSeek V4 Flash结合vLLM实现大规模代码库分析,支持长上下文和多模式推理。

入选理由:DeepSeek V4 Flash支持百万级token上下文窗口,适用于大规模代码库分析。

FeaturedVideo#DeepSeek#vLLM#AI#代码分析英文
Scaling Ray Serve LLM on GKE: Performance without losing the developer experience

Scaling Ray Serve LLM on GKE: Performance without losing the developer experience

Google Cloud Blog675 字 (约 3 分钟)
85

Google 与 Anyscale 合作优化 Ray Serve LLM 在 GKE 上的性能,实现吞吐量提升 5 倍、延迟降低 8 倍。

入选理由:通过 HAProxy 集成,减少代理开销并提升吞吐量。

FeaturedArticle#Ray Serve#GKE#LLM#性能优化#Kubernetes英文
The Infrastructure Behind Making Local LLM Agents Actually Useful

The Infrastructure Behind Making Local LLM Agents Actually Useful

Towards Data Science4379 字 (约 18 分钟)
85

Local LLM agents require infrastructure to overcome slow inference and context overflow, solved via vLLM optimization and structured world state — reducing per-call latency from 15s to under 2s and enabling reproducible scientific workflows.

入选理由:使用vLLM优化推理性能,单次调用耗时从15秒降至2秒内

FeaturedArticle#LLM#Agent#Inference#HPC#Open Source英文
Optimize, deploy, and benchmark an open-source LLM with vLLM

Optimize, deploy, and benchmark an open-source LLM with vLLM

DeepLearning.AI496 字 (约 2 分钟)
82

The course introduces how to use vLLM to efficiently deploy open-source large models, covering techniques like quantization and paged attention.

入选理由:70亿参数大模型需约140GB内存,可能需要多GPU支持单次请求。

FeaturedVideo#vLLM#LLM deployment#AI infrastructure英文
[AINews] Cognition raises $1B in $26B Series D

[AINews] Cognition Raises $1B in $26B Series D

Latent Space2907 字 (约 12 分钟)
78

Cognition closed a $1B Series D at a $26B valuation, becoming the largest remaining independent AI agent lab; ARR projected >$1B by EOY 2026; inference optimization is shifting to architectural improvements—EAGLE 3.1, vLLM, and Qwen3.5 significantly enhance long-context stability and throughput.

入选理由:Cognition D轮融资10亿美元,估值达260亿美元, 成为最大独立AI智能体实验室(2026年5月)

FeaturedArticle#AI Agent#Financing#Inference Optimization#DeepSeek#Cognition英文
Andrew Ng(@AndrewYNg) 图标

New Course on Efficient LLM Serving by Andrew Ng

Andrew Ng(@AndrewYNg)208 字 (约 1 分钟)
75

Efficient LLM serving relies on quantization and vLLM's smart memory management to overcome 140GB VRAM and KV Cache bottlenecks for low-latency concurrency.

入选理由:70B参数模型仅加载权重需约140GB显存,每个活跃请求还需独立KV Cache存储上下文。

FeaturedTweet#LLM Serving#vLLM#Quantization#DeepLearning.AI英文
TokenSpeed is a brand new inference engine purpose built for speed-of-light agentic workloads.  

Re...

TokenSpeed is a new open-source LLM inference engine optimized for agentic workloads, featuring advanced KV caching, an efficient scheduler, and a modular kernel architecture with multi-silicon support.

入选理由:TokenSpeed 实现了媲美 TensorRT-LLM 的性能与接近 vLLM 的易用性。

FeaturedTweet#LLM Inference#NVIDIA#Open Source#KV Cache#Attention Mechanism中英混合
RL post-training is hitting a rollout bottleneck. 

This new paper from #NVIDIAResearch shows how sp...

NVIDIA 研究提出将 speculative decoding 引入 NeMo-RL + vLLM 架构,实现 RL 后训练 rollout 阶段无损加速:8B 模型吞吐提升 1.8 倍,235B 模型端到端预计提速 2.5 倍。

入选理由:RLHF/RLAIF 后训练的 rollout 阶段已成为性能瓶颈

FeaturedTweet#RLHF#speculative decoding#vLLM#NeMo-RL#NVIDIA中英混合
> Ecosystem: Compatible with llama.cpp, MLX, @LMStudio, vLLM, @ollama, @UnslothAI, and SGLang.
&g...

Google AI Developers: Gemma 4 Ecosystem Compatibility and Downloads

Google AI Developers(@googleaidevs)78 字 (约 1 分钟)
65

Google announces its model weights are compatible with major open-source ecosystems and can be directly downloaded from Hugging Face and Kaggle, lowering deployment barriers.

入选理由:Gemma 4 权重与 llama.cpp、vLLM、Ollama 等生态兼容,便于本地部署与推理。

FeaturedTweet#Gemma#Open-source Ecosystem#Model Deployment#Hugging Face#Kaggle英文
前面帖子有人问 vllm 夸Deepseek的博客在哪听。

让 AI 总结小宇宙播客,写了篇文章,其中不少地方提到Deepseek。

“幻方出来的工程师,骨子里就是把系统性能当生死线来对待的”

播客总结文章提及vLLM项目与Deepseek工程师文化,但缺乏深度技术细节和实践指导。

入选理由:vLLM项目由Deepseek工程师推动,强调系统性能优化

FeaturedTweet#vLLM#Deepseek#AI#播客中英混合
PyTorch Conference North America Schedule Is Live

PyTorch Conference North America Schedule Is Live

PyTorch Blog255 字 (约 2 分钟)
60

PyTorch北美会议日程公布,涵盖AI生态、编译器创新及负责任AI等主题。

入选理由:2026年10月20-21日于旧金山举办PyTorch北美会议

FeaturedArticle#PyTorch#AI会议#开源#深度学习英文
Local GenAI on Jetson: OSS models using different inferencing frameworks: Ollama, llama.cpp, & vLLM

文章介绍了在Jetson设备上使用不同框架(如Ollama、llama.cpp、vLLM)部署开源生成AI模型的方法,但内容以视频链接和导航元素为主,缺乏深度技术细节。

入选理由:文章提到Ollama、llama.cpp、vLLM三种框架可用于Jetson设备上的GenAI模型部署。

FeaturedVideo#Jetson#GenAI#Ollama#llama.cpp#vLLM英文
New short course: Fast & Efficient LLM Inference with vLLM, built in partnership with @RedHat and ta...

New Short Course: Fast & Efficient LLM Inference with vLLM

DeepLearning.AI(@DeepLearningAI)168 字 (约 1 分钟)
55

DeepLearning.AI and RedHat launched a free short course teaching open-source model quantization, vLLM deployment, and benchmarking across speed, cost, and accuracy.

入选理由:课程涵盖开源LLM量化技术,直接降低显存占用与推理成本。

FeaturedTweet#vLLM#LLM Inference#Model Quantization#DeepLearning.AI英文
Introducing: Cohere Command A+

We’ve created our most powerful LLM yet, optimized it to run on as l...

Introducing: Cohere Command A+

cohere(@cohere)98 字 (约 1 分钟)
55

Cohere released its most powerful LLM to date, Command A+, optimized to run on minimal hardware and released as open source.

入选理由:Cohere推出最强LLM模型Command A+

FeaturedTweet#Large Language Model#Cohere#Open Source AI#Command#Hugging Face英文
@vllm_project Get started with the code👇 https://t.co/S1cNx6qc8L

@vllm_project Get started with the code👇 https://t.co/S1cNx6qc8L

NVIDIA AI(@NVIDIAAI)203 字 (约 1 分钟)
40

NVIDIA AI 官方账号转发 vLLM 项目启动链接,并附带指向 NVIDIA-NeMo/RL GitHub 仓库的短链,内容无技术细节或上下文。

入选理由:仅含推广性短链接,无代码说明、性能数据或使用指南

FeaturedTweet#vLLM#NVIDIA#LLM inference#GitHub中文

跨材料问答 · vLLM

回答基于:vLLM 相关 30 条材料
    0 / 500

    AI may generate inaccurate information. Please verify important content.