b8327
The popular open-source inference engine corrects a key audio function name in its experimental mtmd module.
The ggml-org team, maintainers of the widely-used llama.cpp project, has released a new commit tagged b8327. This update focuses on a specific correction within the experimental multi-modal (mtmd) module, renaming the function `mtmd_get_audio_bitrate` to `mtmd_get_audio_sample_rate`. The change addresses a fundamental misnomer: the function returns the audio sample rate (e.g., 44.1 kHz), not the bitrate (which is calculated as sample_rate × bit_depth × channels). This aligns the function name with its actual purpose and usage throughout the codebase.
While this is a breaking API change, the developers note it is permissible as the mtmd module is still in an experimental and development phase. The release is accompanied by a full suite of pre-built binaries for developers, covering major platforms and hardware accelerators. This includes builds for Apple Silicon and Intel macOS, iOS, various Linux configurations (CPU, Vulkan, ROCm 7.2), and multiple Windows targets supporting CPU, CUDA 12/13, Vulkan, SYCL, and HIP. The continued expansion of supported backends, like the newly listed openEuler builds, highlights the project's commitment to broad hardware compatibility for efficient, local LLM inference.
- Renames `mtmd_get_audio_bitrate` to `mtmd_get_audio_sample_rate` for accuracy, as it returns sample rate, not bitrate.
- Acknowledged as a breaking change but deemed acceptable due to the mtmd module's experimental status.
- Release includes extensive pre-built binaries for macOS, Linux, Windows, and openEuler across CPU, CUDA, Vulkan, and ROCm backends.
Why It Matters
Maintains code clarity for developers building multi-modal AI applications and ensures the robust, cross-platform llama.cpp engine continues to evolve precisely.