v3.4.0
📦 localaiView on GitHub →
✨ 7 features🐛 1 fixes🔧 5 symbols
Summary
LocalAI 3.4.0 introduces significant expansion with new TTS and AI backends (KittenTTS, kokoro, dia), enhanced WebUI image generation controls, and support for reasoning effort in chat completions. It also improves backend installation flexibility and fixes a llama.cpp rope defaulting issue.
Migration Steps
- If using new backends (KittenTTS, kokoro, dia), be aware that the first call will incur a delay as the model files are downloaded during warm-up.
✨ New Features
- WebUI now allows setting the image generation size.
- Added support for KittenTTS, kokoro, and dia backends; models can be installed directly from the gallery.
- New backends require a warm-up call during the first use to download model files.
- Support for reasoning effort added to OpenAI chat completion endpoint.
- Diffusers backend is now available for l4t images and devices.
- CLI backend installation now supports supplying alias and name overrides (--alias and --name).
- Backends can now be sideloaded by dragging and dropping them into the backends folder.
🐛 Bug Fixes
- Fixed llama.cpp to not default to linear rope.
🔧 Affected Symbols
llama.cppdiffusersKittenTTSkokorodia