llama.cpp b10153 adds support for Nanbeige4.2 model
New release brings community-developed model to the runtime with bug fixes.
ggml-org has shipped llama.cpp b10153, a significant maintenance release that expands model compatibility and tightens code quality. The headline feature is support for Nanbeige4.2 (#25994), a community-developed model now runnable natively on llama.cpp. The update also resolves a flake8 lint check error, corrects a loop bound check, and removes a redundant head_dim calculation, improving both stability and performance.
This release ships binaries for a wide range of targets: macOS (Apple Silicon arm64 with optional KleidiAI, Intel x64, iOS), Linux (x64 and arm64 for CPU, Vulkan, ROCm 7.2, OpenVINO, SYCL FP32/FP16), Android arm64, Windows (x64 CPU, arm64 CPU, OpenCL Adreno, CUDA 12/13, Vulkan, OpenVINO, SYCL, HIP), and openEuler (x86 and aarch64 with ACL Graph). The breadth of platform support underscores llama.cpp's role as a universal local inference engine for AI researchers and developers.
- Adds official support for Nanbeige4.2 model (PR #25994)
- Fixes flake8 lint check, loop bound errors, and redundant head_dim
- Binaries provided for macOS, Linux, Windows, Android, and openEuler across CPU and GPU backends
Why It Matters
Expands llama.cpp's model ecosystem, giving developers more choice for local inference with improved stability.