Laya MPS
github.com/afshinm/laya-mps- Category
- Developer Tools
- Rank
- No. 1571Tools index
Previous survey · No. 1519 ·
- Pricing
- Open Source
- Type
- TOOL
- Use case
- Models: Train & Run
- GitHub
- 26 stars
- Date
About
Laya MPS runs a small typed-decision model locally on Apple Silicon Macs using Metal Performance Shaders, choosing among options, scoring inputs, or estimating whether statements are true rather than generating open-ended text. Its English model is tuned for customer service, invoices, security incidents, and agent traces, running at roughly 32ms median latency using about 2.1GB of RAM (or ~0.74GB in a lower-memory mode) on an M5 Pro.
At a glance
Laya MPS runs the Laya typed-decision model on an Apple Silicon Mac using the GPU. It is worth evaluating if you want to route or score inputs locally and can check whether the model fits your tasks.
Good fit
Mac developers testing local classification under a memory and latency budget.
Consider something else when
Use a generative model if you need free-form text or a different runtime if your deployment is not on a supported Mac.
Start here
Use a small set of synthetic inputs with known labels. Compare accuracy, latency and memory use across the documented modes before choosing one.
Cost and requirements
The project lists Apple Silicon, macOS 14 or newer and about 4 GB free disk space for the runtime, cache and model.
Limits to know
Published timings reflect particular Mac hardware. Evaluate confidence and errors on your own labels and compare performance on the device you plan to use.
Related work
- Laymbda (context)
Sources checked 2026-11-05. Source-based assessment.
Why it made the leaderboard
Lets developers get typed, low-latency decisions like routing or urgency scoring on-device, without paying LLM API latency or cost for narrow classification tasks.
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.