ggerganov/llama.cpp Released Build b12088 with NVFP4 LFM3 GGUF Support
Llama.cpp release b12088 officially adds support for NVFP4 LFM3 GGUF models. This update introduces specific implementation code to handle this low-precision quantization format within the runtime framework. This update expands local LLM optimization by incorporating modern low-precision quantization formats. It enables users to run specialized models more efficiently on compatible hardware architectures. The release is tracked via pull request #31608 and includes precompiled binaries and libraries for a wide range of platforms such as macOS, Linux, Android, and Windows.
## BACKGROUND
GGUF is a binary file format designed to store models for fast loading and inference using GGML-based runtimes like llama.cpp. NVFP4 is a 4-bit floating point format introduced by NVIDIA to deliver efficient and accurate low-precision AI inference tailored for Blackwell Tensor Cores.