What models can do
LocalAI runs models of many kinds. Each page here shows what a kind of model does, how to call it and which backends serve it. The API features are on by default for new users.
| Group | What it covers |
|---|---|
| Text and retrieval | Text generation, constrained grammars, thinking models, embeddings, reranking, moderation, time-series forecasting and benchmarks. |
| Audio and speech | Speech to text, text to speech, sound classification, speaker diarization, audio transforms, voice activity and voice recognition. |
| Images, video and 3D | Image understanding, object detection, face recognition, image and video generation, 3D generation and animation, and motion capture. |
Some of these have an add-on page: Studio, Voice, Face and Talk. See the add-on catalogue.
Also in Models and backends
- Engines: backends that have their own page.
- Train and quantize: fine-tune a model and quantize it to a smaller file.
- Backends and the compatibility table: which backend runs which model.