llama.cpp Fixes kv-cache Overflow in Shared Cells

Core summary

One sentence to understand this update

llama.cpp has released an update fixing a kv-cache issue where shared cells didn't follow the source cache size, causing overflow in shared K/V tensors.

Impact & opportunity

What this could mean

Developers relying on llama.cpp for local model inference will benefit from more stable performance and reliable memory management, reducing potential crashes.