vLLM v0.27.0 发布：支持 Kimi K3、Qwen3.5，性能大幅提升

AI 摘要

vLLM 发布 v0.27.0 版本，包含 561 个提交，来自 242 位贡献者。该版本新增对 Kimi K3、Qwen3.5、K-EXAONE-2.0 等模型的支持，并升级至 PyTorch 2.13.0，深化 FlashAttention 4 集成，优化 DeepSeek-V4 性能，扩展 Model Runner V2 至非生成任务，并引入容错框架和 Rust 前端 gRPC 控制平面。

正文节选

# vLLM v0.27.0 Release Notes ## Highlights This release features 561 commits from 242 contributors (64 new)! * **Kimi K3 support** with a full stack landing in one release: core model files and kernels (#50089, #50000), Python (#50093) and Rust (#50104) frontends, AttnRes kernels (#50090), DeepGEMM support (#50458), compressed-tensors quantized checkpoints (#50500), DSpark AR fusion (#50242), and an option to shard the shared expert instead of replicating it (#50656). * **More new mode

发布时间：2026-08-11 05:18

抓取时间：2026-08-11 05:33

来源机构：vLLM