llama.cpp Fixes Fused QKV Tensor Split for Uneven Head Sizes

Core summary

One sentence to understand this update

llama.cpp release b11176 addresses a bug in tensor split for fused QKV when K/V head sizes are uneven.

Impact & opportunity

What this could mean

This fix ensures more stable and accurate processing of models with uneven K/V head sizes in llama.cpp, improving model compatibility.