llama.cpp v9530 fixes model param propagation bug across platforms
New release patches a critical regression affecting model parameter loading in llama.cpp
Llama.cpp, the widely used C/C++ implementation for running large language models locally, has released version b9530 with a critical fix. The release addresses a regression where CLI model parameters were not being propagated to the inference engine (issue #23847). This bug affected users who passed custom parameters like context window size, batch size, or GPU layers via command-line arguments — those parameters were silently ignored, leading to unexpected model behavior or reduced performance. The fix (PR #23893) ensures all user-specified parameters are correctly applied.
The b9530 release also provides extensive pre-built binaries across major platforms. For macOS, both Apple Silicon (arm64) and Intel (x64) builds are available, with an additional KleidiAI-enabled variant for arm64. Linux users get Ubuntu builds for CPU (x64, arm64, s390x), Vulkan, ROCm 7.2, SYCL FP32, and OpenVINO. Windows binaries cover CPU (x64, arm64), CUDA 12/13, Vulkan, and HIP. Android arm64 is also supported. Developers can download the appropriate package from GitHub releases. This fix is recommended for all users experiencing parameter inconsistencies.
- Fixes model parameter propagation bug (issue #23847) where CLI arguments were ignored
- Pre-built binaries for macOS, Linux, Windows, and Android across multiple backends
- Includes CUDA 12/13, Vulkan, ROCm 7.2, OpenVINO, and KleidiAI support
Why It Matters
Ensures llama.cpp users get consistent LLM behavior by restoring proper parameter handling critical for custom inference setups.