llama.cpp Fixes Vulkan Stale Memory Reuse in Flash Attention and Softmax

Core summary

One sentence to understand this update

llama.cpp version b11414 resolves a critical Vulkan bug concerning stale prealloc_y reuse across flash attention and softmax operations.

Impact & opportunity

What this could mean

This fix improves stability and potentially performance for llama.cpp users leveraging Vulkan, especially on macOS/iOS.