What changed
Five signed llama.cpp prereleases inside the 13 September India boundary extended OpenCL no-shuffle row-alignment handling to q4_K, q5_K and q8_0 quantizations, serialized simultaneous Vulkan queue submissions to avoid an NVIDIA driver synchronization failure, improved SYCL memory-error and build detection, and guarded s390x repack helpers for VXE-capable hardware. The sequence ended at b10946; b10941 only reduced flash-attention test sizes.
Why it matters
The changes target output correctness, crashes, initialization failures and backend availability on specific OpenCL, Vulkan, SYCL and s390x configurations. Operators on those paths have concrete reasons to qualify the completed India-date build rather than assuming a prerelease label or successful startup establishes correctness.
What to do
Pin b10946 and compare it with the last qualified build on the exact models, quantizations, GPU and driver, SYCL toolchain and s390x hardware you operate. Exercise concurrent Vulkan work, affected OpenCL row sizes, memory exhaustion and unsupported build configurations; compare outputs and logs, then prove rollback.
Caveat
Every build is a fast-moving prerelease aimed at narrow configurations. The Vulkan change is a project workaround for an NVIDIA driver problem rather than a root-cause driver fix, and there is no broad independent correctness or performance evaluation for this sequence.