b10536
ggml-org/llama.cppb10536Aug 21, 2026by github-actions[bot]same day as b10534
What changed since b10534
Adds lazy loading for startup models and maintains CUDA decode optimizations.
New features
- Lazy-load startup_models in the server router after main setup.
- Restricts model population to the first load.
Improvements
- Improves initial server startup time by deferring model loading.
Upgrade notes
No configuration changes are required. The feature is backward compatible and improves startup latency by deferring model loads.
Analyzed Oct 17, 2026
Upgrading?
Release notes summary
Adds a lazy-loading feature to the server router that delays the loading of startup models until after the main server setup is complete.
Key Highlights
- Lazy-load startup_models in the server router
- Restricts model population to the first load
- Improves initial startup time
Release Notes (excerpt)
<details open> server: (router) lazy-load startup_models after main setup (#27424) * server: (router) lazy-load startup_models after main setup * only allow is_first_load to populate it * nits * nits 2 </details> **Website:** - <https://llama.app> **Attestations:** - <https://github.com/ggml-org/llama.cpp/attestations/42025993> **macOS/iOS:** - macOS Apple Silicon (arm64) - macOS Apple Silicon (arm64, KleidiAI enabled) DISABLED - macOS Intel (x64) - iOS XCFramework **Linux:**…
Published by ggml-org on GitHub.
Read the full release notesRelated Repositories
:electron: Build cross-platform desktop apps with JavaScript, HTML, and CSS
Godot Engine – Multi-platform 2D and 3D game engine
ClickHouse® is a real-time analytics database management system
Self-hosted game stream host for Moonlight.
DuckDB is a high-performance analytical database system