spec

This release bumps the llama.cpp version to 0.2.0 and ggml to 0.21.0, incorporating numerous backend improvements, bug fixes, and new model support across various hardware accelerators like SYCL, OpenCL, Metal, and Vulkan.

This release introduces auto-detection for MTP draft model types. It also provides pre-compiled binaries for various platforms and hardware configurations, including macOS, Linux, Android, and Windows with support for CPU, Vulkan, OpenVINO, SYCL, ROCm, and CUDA.

This release includes a fix for the server component, specifically addressing the correct handling of accepted tokens during draft token replays. It also provides updated binary downloads for various platforms and hardware configurations.

This release introduces DFlash v2 support, including sliding window attention configuration, and provides updated pre-compiled binaries for numerous platforms and hardware accelerators.

This release introduces spec support for Qwen3.5 and Qwen3.6 via eagle3 and includes various internal fixes and build updates across supported platforms.

This release includes a fix for draft model checkpoints within the spec module and general code clean-up. A warning related to ngram-mod reset is now controlled by the verbose flag.

This release primarily refactors parameter handling within the 'spec' context, including renaming 'sparam' to 'sampling' and adding a new parameter category.

This release focuses on various improvements to the spec module, particularly enhancing the ngram-map functionality and updating related documentation. New binaries are provided across multiple operating systems and hardware configurations.