llama.cpp Just Dropped Hugging Face Presets and Revamped Its Router — Why This Matters
The popular LLM inference engine refines its server routing and drops preset.ini support.
Deep Dive
ggml-org's llama.cpp released b9703, reworking the server router's Hugging Face preset repo handling. This update temporarily removes HF remote preset support, deletes preset.ini, and removes the unused get_remote_preset_whitelist() function. It also prints a warning, adds docs, and removes a stray file. Builds are available for macOS, Linux, Windows, Android, and more, with UI assets updated.
Key Points
- Server router reworked for HF preset repo, removing remote preset support temporarily.
- preset.ini file and get_remote_preset_whitelist() function removed to simplify code.
- Builds available for macOS, Linux, Windows, Android, and iOS across multiple backends.
Why It Matters
Streamlines llama.cpp server, reduces maintenance burden, and improves router reliability for self-hosted LLM deployments.