Preconfigured presets and build/run scripts for llama.cpp and turboquant — GGUF model servers and benchmarks. Two preset files define model configurations with server parameters, quantization, and speculative decoding settings: Each section defines a model (e.g. [Qwen3.6-35B]) with HuggingFace repo, quantization, temperature, and sampling params. Sections ending in -MTP enable speculative decoding via MTP. Results saved to bench-results/. See --help in the script for examples. 1. Clone llama.cpp (or turboquant) to the path expected by the scripts 2. Run the appropriate build-.sh / build-.bat 3. Start the server, e.g../server.sh 4. Hit http://localhost:8080 or use the OpenAI-compatible API…