What's Changed
- [Spec Decode] Report context-limited drafting fallback by @StevenWang-CY in #850
- Speed up the Q6_K qmv and GEMM paths by @ricky-chaoju in #849
- [Perf] Derive the MLA continuation threshold from attention dims by @begcdn in #848
Full Changelog: v0.30.0.dev20260927121512...v0.30.0.dev20260927162153