b8147

The open-source project llama.cpp, maintained by ggml-org, has released a significant server update tagged b8147. This patch specifically addresses GitHub issue #19854, which detailed a bug causing query parameters to be lost when the server's multi-model router proxied requests. For developers using llama.cpp's server to host and route between different AI models (such as various Llama 3 sizes or other GGUF-compatible models), this bug could silently break API integrations, leading to incorrect model loading, failed context passes, or malformed responses. The fix is essential for maintaining the integrity of complex, multi-model AI applications built on this popular local inference engine.

The technical fix involves re-encoding query parameters using `httplib::encode_query_component` within the proxy logic, ensuring all metadata survives the routing process. This update is available across all major platforms, including pre-built binaries for macOS (Apple Silicon/Intel), Linux (CPU, Vulkan, ROCm), and Windows (CPU, CUDA 12/13, Vulkan). For teams deploying llama.cpp in production—especially for agentic workflows, evaluation suites, or A/B testing between models—this patch stabilizes a core infrastructure component. It underscores the project's maturation from a simple model runner to a robust backend for scalable, local AI deployment.