Research radar

Releases, engineering blogs, paper feeds, and conferences worth following.

vLLM releases: A release every few weeks changes defaults, scheduler behaviour, and the supported model list. Read the notes before you upgrade a production engine, not after.

SGLang releases: New disaggregation, router, and kernel features are documented in the release notes first, often before the docs catch up. Check here when a feature you need seems to be missing.

TensorRT-LLM releases: Each stable tag moves the CUDA and driver floor and lists the models and precisions that gained optimized kernels. Filter out the prereleases and re-check before any image bump.