llama.cpp Release b10359 Introduces WebGPU Backend Updates and CI Fixes
Release b10359 of llama.cpp introduces minor continuous integration (CI) fixes and incremental updates for the WebGPU backend. Key updates include adding 32-bit integer (i32) support to copy operations and disabling subgroup matrices when the maximum key-value tile size is zero. These incremental improvements enhance the stability and compatibility of llama.cpp's WebGPU backend, facilitating smoother large language model inference directly in web browsers. It ensures that developers targeting web-based AI deployments experience fewer build failures and better operation coverage. The release resolves CI errors from pull requests #25025 and #25262, tests a new flash attention implementation, and enables the "all ops" test suite. It also provides pre-built binaries for various platforms including macOS, Linux, Windows (with CUDA 12/13, Vulkan, ROCm), and Android.
## BACKGROUND
llama.cpp is a popular open-source software library designed for efficient LLM inference on consumer hardware using the GGML tensor library. WebGPU is a modern web standard and API that allows web applications to access a system's GPU for high-performance graphics and machine learning tasks directly in the browser.