llama.cpp enhances kv-cache efficiency for Gemma-4, improving MTP performance.

Core summary

One sentence to understand this update

The llama.cpp project merged a pull request by ggerganov that optimizes kv-cache by avoiding cell copies, specifically improving MTP performance for Gemma-4.

Impact & opportunity

What this could mean

This enhancement allows for more efficient local deployment and faster inference with Gemma-4 models on llama.cpp, benefiting developers working with quantized LLMs.