Laya on MLX
laya-mlx is an independent MLX port of Laya that runs the original weights on the Apple Silicon GPU with no PyTorch or Transformers runtime. It keeps upstream prompt formatting, calibration and output schema, and it is the base for many Mac demos, agents and benchmarks.
260 projects
Jev-Laya-Demo
johanlrdahl
Demonstrates Laya ticket triage against local Ollama models with typed decisions and quality reports
laya-browser-agent
Sreenuraj
Automates web tasks locally using Laya MLX and Playwright tools
Quantlabs_G-L
rmrodriguezgit
Simulates Polymarket BTC up/down trading using GANs, Q-learning, and Laya MLX audits
laya-local
ira-at-work
Benchmarks local Laya for coding-agent command gating and context-compaction decisions
laya-pokesurvivor
garykhfung
Uses a local Laya checkpoint to control a browser-based PokéSurvivor game on Apple Silicon
laya-api
snorripall
Provides an API for running the Laya model on Nvidia and Apple Silicon hardware
laya-mobilecli
nichotined
Resolves plain-English mobile UI tap targets with the Laya typed-decision model
flylaya
leewbl
A fruit-fly behavior study combines connectome-driven motion with a Laya intent layer
laya-lab
karth2512
Benchmarks Laya classification and fine-tuning methods against classical baselines and the Jev API
sys1_mcp
dmonteroh
Provides an MCP server and HTTP daemon for System One decision models, including an in-process Laya backend
laya-decision-lab
geomichelon
Runs local Laya inference for ticket triage with typed decisions, policy review, and sample evaluation
pi-laya-mlx
a3harva
Screens coding-agent tool calls with a fine-tuned Laya model running locally on Apple Silicon
laya-docker
thomashaertel
Runs Laya on AMD GPUs in Docker and exposes Jev-compatible decision APIs and an MCP server
laya-fastapi
builtbybasit
Provides a local Jev-compatible API powered by Laya with MLX and PyTorch backends
laya-oneshot
gergeisabo
Runs one typed decision with Laya on demand, then exits without keeping a server running
trend-agent-webapp
Lucas-0911
Scores media trends with Jev or local Laya-MLX and feeds the results into an analysis workflow
pmax
will702
A Go Mac productivity CLI optionally uses an MLX/Laya sidecar for semantic search and improvement
jev-vs-laya
KPMGE
Compares Laya base and fine-tuned models with hosted Jev for support-message triage
laya-zcode-stack
Ydjin1984
Runs a local Laya decision engine and ZCode plugin with MCP tools for coding-agent workflows
laya-browser-grounding
Who-Code
A study and demo using Laya-MLX to ground browser automation actions on page elements
Install
Requirements from the laya-mlx README: Apple Silicon, Python 3.11+, macOS 14+.
pip install laya-mlx
Pre-converted FP16 checkpoints are on Hugging Face: aac6fef/laya-mlx, aac6fef/laya-multilingual-mlx and aac6fef/laya-typed-decisions-mlx. You can also load the original convaiinnovations/laya IDs.
Router and CLI
The port includes the upstream language router and presets:
from laya_mlx import Router, triage_questions
router = Router(dtype="float16", max_loaded=2)
result = router.predict({"message": "发票被重复扣款，请退款。"}, triage_questions())
print(result["routing"]) # multilingual
For the terminal demo, where Laya plays Snake:
pip install 'laya-mlx[demo]'
hf download aac6fef/laya-multilingual-mlx
laya-snake
Reported numbers
On an M3 Max, FP16, one short question: 13.42 ms P50 for the 421M English model and 7.39 ms for the 322M multilingual model. The README states that all three checkpoints matched the upstream answer on 63/63 validation questions in FP32 and FP16. That measures port fidelity, not task accuracy.
Projects built on laya-mlx
- laya-ultrafast: a port of browser-use/jev-ultrafast that makes browser decisions locally.
- laya-mcp (wsargent): an MCP server with triage, guard, moderation and email tools.
- laya-fast: adds Neural Engine execution next to the MLX GPU.
- laya-jev-lab and jev-laya-benchmark: independent Jev comparisons that use laya-mlx as the Laya side.
- laya-vs-jev: Laya and Jev playing Chrome's T-Rex game side by side.
Caveats
This is not an official Convai Innovations release. Training and fine-tuning stay in the upstream project. BF16 can be requested but is not part of the published validation.
More ways to use Laya