🚨Meetup Alert🚨 Scaling LLM inference? Join us June 10 in SF for a Ray Meetup with real-world wins from Pinterest + Anyscale We'll discuss: ⚡ 300x throughput w/ Ray at Pinterest ⚡ DeepSeek + vLLM prod deployment ⚡ Ray Serve + Data LLM preview 📍 Anyscale HQ |

Excited to open-source Shift Parallelism, developed at Snowflake AI Research for LLM inference! With it, Arctic Inference + vLLM delivers: 🚀3.4x faster e2e latency & 1.06x higher throughput 🚀1.7x faster generation & 2.25x lower response time 🚀16x higher throughput

vLLM 0.9.0 is HERE, unleashing HUGE performance on AMD GPUs! MI-Series: FP8 KV cache, +19.4% with AITER GEMM, +16.8% Qwen3 MoE (MI300X), +13.8% DeepSeek V3/R1! Consumer: 📈 RX 9000 Series: +16.4% throughput 🚀 RX 7000 Series: +19.0% performance gains AI at AMD AMD Radeon

🚀 Join us at the SF AIBrix & vLLM Meetup on June 18th at AWS SF GenAI Loft! Learn from experts at ByteDance, AWS Neuron, and EKS. Discover AIBrix: a scalable, cost-effective control plane for vLLM. Talks, Q&A, pizza, and networking! 🍕🤝 lu.ma/ab2id296

.vLLM v0.9.0 was a BIG release 🎉 📝 649 commits 👥 143 contributors 👏 82 first-time contributors Huge thanks to everyone who made it happen! Michael Goin breaks down what’s new in vLLM v0.9.0 ⬇️

🇯🇵 Join us for an in-person vLLM meetup on Monday, June 16 in Tokyo. Or tune in via live stream! Agenda: -Intro to vLLM -Japanese LLM adoption -Model optimization w/ LLM Compressor -Distributed inference w/ llm-d -Q&A and lightning talks RSVP: ossbyredhat.connpass.com/event/357695/

Congrats on the launch! vLLM is proud to support the new Qwen3 embedding models, check it out 👉🏻 github.com/QwenLM/Qwen3-E…

Thanks for the great investigation! vLLM values usability, performance, and building the ecosystem for LLM inference, together let's make open-source better❤️ Stay tuned for latest updates from vLLM!

Our next Ray Meetup is almost here! June 10th 👇 Hear how Pinterest scaled inference 300x with Ray, watch a DeepSeek + vLLM live demo, and get a first look at new Ray Serve tools. Plus: networking, snacks, and great convos with fellow builders. Save you seat:

⬆️ uv pip install -U vllm --extra-index-url wheels.vllm.ai/0.9.1rc1 --torch-backend=auto Try out Magistral on with vLLM 0.9.1rc1 today! 🔮

👀 Look what just arrived at UC Berkeley Sky! 🌟 A shiny MI355X system. Huge thanks to AMD for supporting open source and we are looking forward to getting it set up in the next few days!

Glad to support the UC Berkeley Sky and the vLLM community. Day-0 Support means you get hardware on Day -2 😀. Looking forward to what the community builds and accelerating AI adoption.

This table was a footnote at the end of the blog, but it's actually one of the most interesting points. There is an emerging stack for post-training. anyscale.com/blog/ai-comput…

Thank you AMD Lisa Su Anush Elangovan for Advancing AI together with vLLM! We look forward to the continued partnership and pushing the boundary of inference.

Cool to see vLLM used as part of WhatsApp Trusted Execution Environment (TEE) Private Processing! Paper here: ai.meta.com/static-resourc…

vLLM AMD Anush Elangovan You’re very welcome! Looking forward to all vLLM will do to enable the community!

🇸🇬!

vLLM now recommends uv's `--torch-backend=auto` for automatic CUDA selection at install-time