llama.cpp Release b10907 Fixes KV Cache for DeepSeek2, GLM4MoE

Core summary

One sentence to understand this update

llama.cpp version b10907 was released, primarily addressing fixes for MTP context KV cache allocation in models like DeepSeek2, GLM4MoE, and Cohere2MoE.

Impact & opportunity

What this could mean

Users of llama.cpp who work with these specific large language models can expect improved stability and performance due to the KV cache fixes.