What is mattwood.fyi?

A live, structured feed of riffs and links from Matt Wood, Chief AI & Technology Officer at AWS — readable by people and queryable by agents as a typed knowledge graph.

Is there an API and is it free?

Yes. A public, read-only query API is available with no authentication and no API keys. See https://mattwood.fyi/openapi.json and https://mattwood.fyi/docs.

How should an AI agent use this site?

Fetch https://mattwood.fyi/feed.json for the full structured feed, or query the API for keyword/semantic search and the connection graph. See https://mattwood.fyi/agents/ for when-to-use guidance.

vLLM v0.28.0 Release — mattwood.fyi

Connections

Supports

A simple fix for LLM tail latency vLLM is a primary tool for optimizing LLM inference latency; its new release directly addresses tail latency improvements mentioned in the existing item
Trismik: Evidence-Based Model Switching vLLM's improved inference capabilities enable evidence-based model switching tools like Trismik by making multi-model serving more efficient and cost-effective

Related to

AMD acquires AI chip startup Taalas to boost inference performance by etching models into silicon AMD's acquisition of Taalas for silicon-level model inference and vLLM's software-level inference optimization represent converging hardware/software approaches to the same LLM inference performance challenge
Nemotron 3.5 Lightning and NeMo Switchyard for Agentic AI Nemotron and NeMo Switchyard for agentic AI and vLLM both target high-performance LLM serving infrastructure, with overlapping use cases in agentic deployments
Fireworks AI: Specialized Intelligence Infrastructure Both concern specialized LLM inference infrastructure — vLLM as an open-source library and Fireworks AI as a commercial inference service compete/complement in the same space
MiniMax H3 Inference Engine for Mac Both MiniMax H3 and vLLM are inference engines targeting efficient LLM serving, representing parallel approaches to the same optimization problem

Introducing System One Models and Jev Both are infrastructure-level AI platform releases (vLLM v0.28.0 and System One Models) targeting developers who need to deploy and serve models at scale
Mercury 2.5 Release Both are framework/tool version releases announcing new features and improvements to existing software infrastructure
Harness Engineering Paper Collection vLLM is a prominent inference and serving harness framework; its release notes reflect the ongoing engineering of sampling and execution infrastructure that the paper collection traces from 2019 onward
OpenCodex: Universal LLM Provider Proxy Both are infrastructure-layer tools for LLM deployment flexibility — vLLM provides high-performance serving while OpenCodex provides a compatibility proxy layer; together they address the LLM accessibility stack

Supported by

Qwen3.8 27B Quantization Benchmarks vLLM is a primary inference engine that leverages quantization techniques; benchmark data on optimal quantization levels (4-bit sweet spot) directly informs vLLM deployment configurations
Declarative Attention: Models Controlling Their Own Attention vLLM is a major inference serving framework; Declarative Attention's KV cache reduction could be implemented within or benchmarked against vLLM's caching infrastructure

Type:	link
First published:	Tuesday, 1 September 2026 at 20:39 UTC
Location:	https://github.com/vllm-project/vllm/releases/tag/v0.28.0
URL:	https://mattwood.fyi/i/f9631342

I'm Matt Wood, and this is For Your Information. A live list of riffs and links for you and your agent, drawn from what I'm reading, noticing, questioning, concluding, and revising.

Connections