OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Infrastructure & Serving

Laya Mlx

Laya Mlx is an Apache-2.0-licensed Python library that runs Laya typed-decision models locally on Apple Silicon using MLX. It produces probabilities for named choices, rubric scores and true-or-false propositions without generating text tokens or calling a cloud API.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Laya Mlx — OpenViST project profile in AI Infrastructure & Serving.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use it when an Apple Silicon application needs local, constrained decisions rather than free-form text generation, and its typed outputs fit the task.

Who it is for

  • Python developers building local decision features into macOS applications on Apple Silicon.
  • Teams that need typed outputs such as option probabilities, rubric scores or proposition probabilities.
  • Developers exploring multilingual typed-decision inference or decision-driven interactive applications.

Documented uses

  • Route a message among named categories, such as billing, technical or sales.
  • Assess a state against ordered rubric levels and obtain an expected score.
  • Estimate whether a proposition is true.
  • Use model decisions to control an interactive application such as the documented Snake demo.
Before you choose

Requirements & limits

The primary inference workflow uses a Python package and downloads a model checkpoint on first load; later inference is local. Requires Apple Silicon, macOS 14 or later and Python 3.11 or later.

  • This is a runtime and library, not a hosted inference API; applications integrate it through Python.
  • The three documented checkpoints differ in language purpose, parameter count and context limit: 512 tokens for English Laya, 1,024 for multilingual and typed-decisions models.
  • Context limits include state, instructions and options, so the full decision input must fit.
  • Reported M3 Max latency is for short questions and excludes model loading; hardware and workload affect results.
  • The repository covers inference and conversion, while training and fine-tuning remain in the upstream project.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Install the library and load a decision model

Set up the Python package and load a checkpoint for a typed choice prediction.

Before you start

  • Apple Silicon
  • macOS 14 or later
  • Python 3.11 or later
  • Internet access for the first checkpoint download

Steps

  1. Install the package with the command below.

    Documented command
    pip install laya-mlx

  2. Import laya_mlx in Python and load aac6fef/laya-mlx with laya.load(...).

  3. Call agent.predict with a state and a choice question containing instructions and named criteria; read the selected answer from the returned answers object.

Expected outcome

A locally computed typed choice, such as a department assignment for a customer message.

Guide limitations

  • The first model load downloads the checkpoint; later inference runs locally.
  • This recipe covers inference, not training or fine-tuning.
  • The example uses the English checkpoint; model context limits include state, instructions and options.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
10 October 2026
Original website title
mizorewww/laya-mlx · Official homepage title · retained in its source language
How we review sourcesSuggest a correction