logo
Kurashizu Blogwhere ideas flow
  • Home
  • News
  • Blog
  • About
  • Home
  • News
  • Blog
  • About

© 2026 Kurashizu. All rights reserved.

Admin·New Post·Service Status
Powered by Cloudflare·30.10.26
← Back to news

Laya (OS Jev) on Mac M4 CoreML Offline (45 decisions per second)

Original article
gist.github.com|45 points|6 comments|by putna|Sep 20, 2026

Laya (OS Jev) Implementation on Mac M4 via CoreML (Offline)

This report documents the successful offline deployment of Laya on Apple's M4 silicon using CoreML, achieving a high-performance throughput of ≈45\approx 45≈45 decisions per second.

🚀 Quick Start Installation

To replicate this environment, follow these steps using the uv package manager.

Setup Checklist

  • Create project directory
  • Initialize environment
  • Install laya-coreml with demo dependencies
  • Download the ANE-optimized model
  • Execute the snake model

Deployment Commands

# Initialize the workspace
mkdir test-laya && cd test-laya
uv init

# Install the necessary library
uv add 'laya-coreml[demo]'

# Fetch the multilingual CoreML model from HuggingFace
hf download aac6fef/laya-multilingual-coreml-ane --local-dir models/snake

# Launch the model
uv run laya-coreml-snake --model models/snake

📊 Technical Performance Analysis

The system was tested on a cutting-edge ARM64 architecture. Below is the detailed resource utilization captured during sampling.

System Specifications

MetricValue
HardwareMac M4 (ARM64)
OS VersionmacOS 27.0 (Build 26A428)
Python Versionpython3.12
Physical Footprint560.4 MB560.4\text{ MB}560.4 MB
Peak Memory Usage778.0 MB778.0\text{ MB}778.0 MB
Timestamp2026-09-20 23:05:12

Execution Flow


🌐 API Integration & Testing

The author integrated the model behind a Cloudflare typesafe/jev API to facilitate remote inference.

"I just wanted to see what Laya does and put it behind Cloudflare's typesafe/jev API" — fordnox

Sample Request

To test the urgency detection, the following curl command was utilized:

curl https://laya.inference.zaitlabs.com/ai/run \
 -H "Content-Type: application/json" \
 -d '{
   "state": "Help! My payouts have been failing for 3 days.",
   "questions": {
     "is_urgent": {"type": "noul", "instructions": "Does this convey urgency?"}
   }
 }'

Sample Response

The model returned a numerical probability score for the noul type: { "answers" :{ "is_urgent" :{ "type" : " noul " , "noul" : 0.7894 }}}

The urgency score can be represented as: P(urgent)=0.7894P(\text{urgent}) = 0.7894P(urgent)=0.7894


🖼️ Visual Evidence & References

Loading... Uh oh!

Author Profile System Screenshot

Useful Links:

  • Source Code: mizorewww/laya-coreml
  • Terms of Service: laya.inference.zaitlabs.com/terms