Unofficial gallery

100 public builds,
one open model.

A short catalog of projects using convaiinnovations/laya. Browse by the machine it runs on. Every card leaves this site.
Learn a pathBrowse the list
Laya on the MLX port playing Snake, with move probabilities for each turn
Recorded runNext LEFT · 34.8 ms · 11.4/s

UP

DOWN

LEFT

RIGHT

MLX Snake on Apple SiliconOpen this build

choice, score, noul
Typed answers. No paragraph to parse.
421M · 322M · 421M
English, multilingual, and the specialist checkpoint.
~33 ms
ConvAI’s published time for one multilingual question on a T4.

Newly seen

Public posts the crawler picked up. Cards only appear after a strong match on GitHub, Hugging Face, or PyPI, or after inbox review. Last scan 7 Nov. Today’s watch.

No public posts in the last scan. The next pass is scheduled.

On-deviceApple Silicon

Laya MLX

sjoerdbodbijl/laya-mlx

sjoerdbodbijl/laya-mlx · text-classification

sjoerdbodbijlHugging Face
On-deviceGPU

Laya ONNX

kallebysantos/laya-onnx

kallebysantos/laya-onnx on Hugging Face.

kallebysantosHugging Face
On-deviceGPU

Laya ONNX

onnx-community/laya-ONNX

onnx-community/laya-ONNX · text-classification

onnx-communityHugging Face
On-deviceGPU

Laya ONNX

distinctinteractive/laya-onnx

distinctinteractive/laya-onnx on Hugging Face.

distinctinteractiveHugging Face
GamesApple Silicon

75.40Moves/s

Terminal Snake on the MLX port

A local Snake loop for Apple Silicon. Each turn asks the MLX build for a direction, and a safety layer can reject a move that would end the run. The repo’s optimized max-speed note reports 75.40 moves per second across 2,400 moves on an M3 Max.

mizorewwwGitHub
On-deviceNeural Engine

~5 msShort decision

Core ML runtime, including a Snake demo

An independent Core ML port of the typed-decision API, with a terminal Snake demo and a short-question path aimed at the Apple Neural Engine. The project describes short decisions at about 5 ms on an M3 Max.

mizorewwwGitHub
BenchmarksApple Silicon

MLX latency notes

The timing write-up that sits next to the MLX runtime. Treat it as the author’s measurements of that port, not a score we re-ran.

mizorewwwGitHub
BenchmarksNeural Engine

4.98 msP50

Neural Engine speed and energy notes

A published comparison of a short multilingual question on an M3 Max. The note reports 4.98 ms p50 for ANE FP16, and about 2.78× lower whole-system energy per decision than compiled MLX FP16 in that same test.

mizorewwwGitHub
On-deviceApple Silicon

13.4 msEnglish median

laya-mlx package

The installable MLX runtime. Its page reports a 13.4 ms median for a short English decision and 7.4 ms for the multilingual checkpoint, with no generated tokens.

mizorewwwPyPI
On-deviceNeural Engine

4.98 msP50

laya-coreml package

The installable Core ML package. The release note reports 4.98 ms p50 and 5.31 ms p95 for one short multilingual question on M3 Max ANE FP16.

mizorewwwPyPI
On-deviceApple Silicon

63/63Argmax match

MLX weights for the multilingual checkpoint

Apple Silicon weights for the 322M multilingual checkpoint. The card reports that FP16 matched the upstream argmax on 63 of 63 decision distributions in the author’s local check.

aac6fefHugging Face
On-deviceNeural Engine

English 421MSource

Core ML bundle of the English checkpoint

A portable Core ML bundle that scores choice, score, and noul questions on Apple Silicon without a PyTorch runtime. The card marks it as an independent conversion.

aac6fefHugging Face
On-deviceBrowser

63/63Answer match

ONNX export for servers and the browser

A float16 ONNX graph of the multilingual checkpoint, including a path for onnxruntime-web. The card reports 63/63 selected-answer agreement with the MLX runtime on its parity fixtures.

mizchiHugging Face
GamesBrowser

Browser Snake page

The direct Snake page published from the browser demo. Moves come from the in-browser multilingual graph rather than a cloud generate call.

mizchiHugging Face
GamesBrowser

Browser chess page

The direct chess page from the same browser demo. It is a separate board from the Snake page, still using the published ONNX export.

mizchiHugging Face
On-deviceApple Silicon

7–14 msShort decision

mizchi’s MLX runtime fork

A fork of the MLX runtime whose description reports short decisions in the 7–14 ms range on an M3 Max, and which publishes the ONNX export used by the browser demo.

mizchiGitHub
GamesApple Silicon

PyTorch MPS runtime with a Pong demo

A Metal runtime for Macs that keeps inference on the machine. The repo ships a local Pong demo and more than one memory setting for the same typed-decisions checkpoint.

afshinmGitHub
Fine-tunesGPU

0.766Accuracy

Upstream typed-decisions checkpoint

ConvAI’s own specialist, fine-tuned on four synthetic workflows. The card publishes 0.766 accuracy on 2,000 decisions and says the base checkpoints sit near chance on that same set until they are fine-tuned.

ConvAI InnovationsHugging Face
Fine-tunesGPU

Kaggle 2×T4Notebook

Training code and the fine-tune notebook

The upstream repository. ConvAI points to a Kaggle notebook, laya_finetune_typed_decisions_2xT4_kaggle.ipynb, for reproducing the specialist checkpoint on free 2×T4 GPUs.

ConvAI InnovationsGitHub
BenchmarksGPU

32.8 ms1 question

Published speed table on the model card

The hub card for all three checkpoints. ConvAI reports 32.8 ms for one multilingual question on a T4, and a slower English checkpoint when the script is not Latin.

ConvAI InnovationsHugging Face
BenchmarksGPU

322MParams

Multilingual checkpoint card

The 322M checkpoint ConvAI recommends outside English. Their 51-language sweep is the source for the claim that the English checkpoint can stay confident while reading a script it cannot handle.

ConvAI InnovationsHugging Face
BenchmarksBrowser

Official workflow demo

ConvAI’s live Space for trying the published workflows and the language router. It is the authors’ demo, not a third-party game.

ConvAI InnovationsHugging Face

Every figure is the linked author’s reported number. Nothing here was re-run by this site.