llama.cpp Release b9934 Improves ggml-webgpu Flash Attention

Core summary

One sentence to understand this update

llama.cpp's latest release, b9934, includes a tuning for subgroup split in flash_attn_vec for ggml-webgpu, aiming to optimize performance.

Impact & opportunity

What this could mean

This optimization can lead to faster inference for LLMs on WebGPU-compatible devices, providing builders with better performance for local AI applications.