llama.cpp Release b12028 Adds Gemma 4 Architecture Support
llama.cpp release b12028 introduces architecture support and testing capabilities for Gemma 4 within its test-llama-archs suite alongside standard cross-platform binary builds. The update specifically configures the testing framework to utilize Grouped-Query Attention (GQA) for Gemma 4. This update ensures that developers can properly test and validate Gemma 4 models using the industry-standard local inference library. Maintaining rapid architectural compatibility allows the open-source community to efficiently run emerging models on consumer hardware. The pull request specifically enables Gemma 4 validation in the `test-llama-archs` suite by incorporating GQA handling. Standard release packages are provided across multiple platforms including macOS, Linux (with various CUDA and Vulkan backends), Android, and Windows.
## BACKGROUND
llama.cpp is a popular open-source library that enables efficient inference of large language models in C and C++. Grouped-Query Attention (GQA) is an attention mechanism optimization that reduces memory bandwidth requirements during inference, helping models run faster.