Llama.cpp adds auto-detect MTP draft model support
Llama.cpp v1.27.0 ships with auto-detect mtp draft model type.
Deep Dive
llama.cpp release b10415, published 13 Aug by ggml-org, adds a spec
Key Points
- Llama.cpp v1.27.0 introduces auto-detect mtp draft model type (#27005)
- Supports 25 CPU/GPU backends: Apple Silicon, CUDA 12/13, ROCm 7.14, Vulkan, OpenVINO, SYCL
- CUDA builds include 12.4, 13.3 and preview 13.4 DLLs for Windows
Why It Matters
Brings one-click optimized LLM inference across every major CPU and GPU, slashing setup time for developers.