ggerganov/llama.cpp Released Build b12223 with Granite NVFP4 Support
Llama.cpp release b12223 adds support for attaching Granite NVFP4 weight scales and passes NVFP4 scales through the feed-forward network build process for Granite architectures. This release expands local execution capabilities for specialized model weight formats, enabling practitioners to run quantized models efficiently on consumer or edge hardware. The update specifically targets Granite model architectures under the 'granite-nvfp4-moe-scales' branch and was contributed by developers at IBM.
## BACKGROUND
Llama.cpp is an open-source software library designed to enable efficient inference of large language models on consumer-grade hardware with minimal setup. Quantization reduces model size and memory requirements by converting high-precision weights into lower-bit formats, such as NVFP4.