vLLM 0.28.0 lands a big-model serving push for Kimi-K3 and DeepSeek V4

vLLM 0.28.0 (584 commits from 270 contributors, 76 new) is dominated by a Kimi-K3 performance push: Decode Context Parallel support, fused FlashKDA decode/prefill kernels, combined all-gathers with 1.5-3x kernel-level speedup, an adaptive speculative token budget delivering ~60% better DSpark TTFT, and optional shared-expert sharding saving ~17 GiB per GPU. DeepSeek V4 gets sparse MLA working end-to-end for plain decode, MTP and DSpark speculative decoding, plus AMD Quark NVFP4 and ROCm enablement. Other headline items: DFlash2 speculative decoding with a candidate selector, Model Runner V2 maturation (E/P/D disaggregation, weight offloading), tiered KV-cache offloading including disk, a Rust frontend with gRPC (protobuf schemas on Buf), and new defaults (max_num_batched_tokens 8192->16384, Mamba prefix caching on by default). Breaking changes: bitsandbytes support moved to an out-of-tree plugin, Transformers bumped to 5.15.0, deprecated APIs removed.

If you serve Kimi-K3 or DeepSeek V4 weights, this release is where the material throughput and memory wins land — plan an upgrade pass. Everyone else should note the two structural signals: open-weight frontier models now get first-class, model-specific kernel work in the default serving stack within days of release, and bitsandbytes users must install the out-of-tree plugin before upgrading.