vLLM releases: v0.28.0

TL;DR

vLLM released version 0.28.0 with significant optimizations, including Decode Context Parallel (DCP) support and fused FlashKDA decode and prefill kernels, benefiting from 584 commits by 270 contributors. This update enhances performance, particularly for Kimi-K3, highlighting the community's contribution to the project.

Detailed Summary

vLLM released version 0.28.0 with significant contributions from 270 individuals, including 76 new contributors. Key highlights include major optimizations for Kimi-K3, such as Decode Context Parallel (DCP) support and fused FlashKDA decode and prefill kernels. This release underscores the collaborative development process and technical advancements in vLLM's performance.

Key Points

• • v0.28.0 includes 584 commits from 270 contributors.
• • Kimi-K3 performance was significantly optimized.
• • Decode Context Parallel (DCP) support was added.
• • Fused FlashKDA decode and prefill kernels were introduced.