English–Spanish tool-routing models: a 421M teacher, a distilled 150M student, and an MLX release for Apple Silicon.
AI & ML interests
On-device AI, edge inference, small language models, tool routing, MLX, multimodal models, model quantization
Recent Activity
Organization Card
💎 Lapis Labs
Lapis is a private AI app for iPhone, iPad and Mac that runs models directly on your device.
Our goal is simple: build practical, efficient models for personal AI that can work locally, with low latency, low resource usage and explicit control over what the assistant can do.
Website · Download on the App Store
Lapis SAL
Lapis SAL — Semantic Action Layer is our first model family.
It is designed for lightweight tool routing: understanding user intent and deciding which tool should be used, or whether no tool is needed at all.
The models are bilingual in English and Spanish and are optimized for small personal agents and local inference.
| Model | Purpose |
|---|---|
| Lapis SAL 150M | Distilled student model |
| Lapis SAL 421M | Teacher model used for distillation |
| Lapis SAL 150M MLX | MLX release for Apple Silicon |
Supported routing targets include:
Calculator · Calendar · Clock · Code · Contacts · Health · Location · Mail · Maps · Messages · Reminders · Weather · Web Search · NONE
Evaluation
On our private synthetic held-out benchmark of 1,000 English–Spanish examples:
- Teacher: 96.0%
- MLX student: 95.1%
These results are specific to this benchmark and may not reflect real-world performance.
Open models
Our releases are intended for research, experimentation and local AI development.
English & Spanish · Tool routing · Knowledge distillation · MLX · Apache 2.0
datasets 0
None public yet