llama.cpp

Vulkan: Default to 1GB allocations instead of 4GB to avoid fragmentation and driver issues

#12434

Merged