Signals · AI Tech Radar

Titles and summaries are in English where the source is; the editors' notes and signals from Chinese sources are in Chinese.

Tool · 2026-09-22Worth tracking

v0.30.0

v0.30.0 Highlights This release features 762 commits from 315 contributors (104 new)! New models : DeepSeek-V4.1-Flash ( #56214 , #56228 , #56208 ) with the whole KV stored in MXFP8 through the FlashMLA V4.1 record on SM100 ( #56893 ), DeepGEMM Mega-mHC ( #56…

Why it matters自托管推理的成本不只在每个 token，也在每次重启、每次扩容、每次显存不够时。这一版同时动了三处：重启不再重载权重、KV 可以溢出到主机内存、引擎初始化快了三倍多；并把文本水印变成了开源引擎的通用功能。另有六项破坏性变更，升级前要逐条核对。

For 维护自托管 vLLM 服务、负责升级与扩容的推理平台团队、需要为生成内容提供来源证明的合规与平台团队

Tool · 2026-08-10Worth tracking

vLLM v0.27.0

vLLM v0.27.0 ships 561 commits from 242 contributors, landing full-stack Kimi K3 support in a single release, upgrading to PyTorch 2.13, and extending Model Runner V2 to non-generative workloads.

Why it matters它回答了本站 07-16 发布 Kimi K3 时留下的悬念：开放权重公布之后，自托管推理栈多久能落到「有性能路径」那一级。答案是不到一个月，而且是一个版本内落齐。

For 自托管大模型推理服务的平台工程团队、需要判断一次推理栈升级值不值得排期的技术负责人

Tool · 2026-07-25Worth tracking

vLLM v0.26.0

vLLM v0.26.0 ships 411 commits from 212 contributors, headlined by a full support stack for the new Inkling model family.

Why it matters开源万亿参数模型从「放出权重」到「主流推理引擎上可用」的间隔被压到了十天量级，而且是基础建模、图捕获、注意力核、推测解码一次配齐。这改变了自建推理团队评估新模型的排期假设。

For 自建或自运维推理服务的平台团队、需要在本地评估新开源模型的算法与评测团队

Tool · 2026-07-11Worth tracking

vLLM v0.25.0

vLLM v0.25.0 ships 558 commits from 232 contributors. Model Runner V2 becomes the default for all dense models, building on the quantized-model support introduced in the previous release.

Why it matters推理引擎的默认执行栈切换会直接改变吞吐、延迟和 GPU 成本曲线。vLLM 是目前自建大模型服务最主流的开源选择，它的大版本升级几乎影响所有自托管推理团队。

For 推理平台与服务工程师、关注 GPU 成本的 MLOps 负责人

Tool · 2026-06-30Worth tracking

vLLM v0.24.0

vLLM, the open-source high-throughput LLM inference and serving engine, shipped stable release v0.24.0 while the v0.25 release-candidate series is already in flight.

Why it matters自托管推理的事实标准级引擎保持高频迭代：稳定版与下一版 rc 并行推进，意味着私有化部署的性能与成本基线还在持续被刷新。

For 推理平台工程师、需要私有化部署的团队