b12597
The latest release of llama.cpp (v1.25.9) includes a fix for the Qwen3-ASR converter, which now reads the chat template from models/templates instead of embedding its own copy, ensuring consistency across GGUF and test-chat. The release also provides updated binaries for macOS, Linux, Windows, Android, and openEuler platforms with support for various backends including CUDA, Vulkan, ROCm, OpenVINO, SYCL, and Snapdragon hardware.



