llama.cpp b10080 fixes server crash and UI validation bugs
Two critical fixes improve stability and reliability for local LLM inference.
The latest release of llama.cpp (b10080) tackles two reliability issues. First, a server-side bug caused crashes (std::bad_function_call) when the X-Conversation-Id header was present and request validation failed. The error response was upgraded from a generic 500 to a proper 400, and the session is now cleaned up correctly—preventing lingering empty sessions that could interfere with future requests.
Second, a UI fix prevents the backend_sampling setting from sending a placeholder value on fresh profiles. Previously, the placeholder was sent verbatim, causing validation failures for all default setting completions. Now a proper guard filters it, while still allowing explicit true/false values. The release includes pre-built binaries for CPU, CUDA, Vulkan, ROCm, OpenVINO, SYCL, and more across macOS, Linux, Windows, and Android.
- Server now returns 400 (not 500) on validation errors with X-Conversation-Id, preventing crashes and session leaks.
- UI fix filters the backend_sampling placeholder string, allowing default settings to validate correctly.
- Supports numerous platforms: macOS (Intel/Apple Silicon), Linux (x64/arm64 with CPU/Vulkan/ROCm/OpenVINO/SYCL), Windows (CPU/CUDA/Vulkan/HIP), and Android arm64.
Why It Matters
Improves stability and reliability for self-hosted LLM servers and UI, reducing crashes and configuration errors.