vLLM adds an experimental plugin to split attention and FFN serving
vllm_project · x · 2026-07-24
vLLM introduced an experimental AFD Plugin that implements Attention-FFN disaggregation for MoE serving.
- Attention and expert/FFN workloads have different scaling needs, so the plugin splits them into separate services instead of forcing one shared topology.
- The design keeps the same vLLM serving surface and avoids a fork.
- It targets both NVIDIA GPUs and Ascend NPUs.
The attached architecture diagram shows entry points for OpenAI-compatible APIs and vLLM serve, plus separate attention, AFD connector, and FFN roles.
Related event: vLLM Releases Experimental AFD Plugin to Boost MoE Inference(3 posts)→
More from Infra
- 12 KV Cache Reduction Techniques Every AI Engineer Should Understand, Explained — blaizedsouza · 2026-09-11
- The shadow GPU capacity market is formalizing, with Meta selling excess compute to outside buyers — DavidLinthicum · 2026-09-11
- Engram's random reads don't suit SSDs; CPU-memory over NVLink could serve all 72 GPUs — bookwormengr · 2026-09-11
- 80% of the DIY LLM inference hype posters have already quit — it's brutally hard systems work — abhijithneil · 2026-09-11
- Hugging Face's Ultra Scale Playbook: a free book on training LLMs on GPU clusters — mdancho84 · 2026-09-11
- Is inference latency becoming the biggest bottleneck for production AI agents? — Euphoric_Sea632 · 2026-09-11