Skip to content
Torad. LAB / 001
Talk to us
Torad Labs runs offline

We train small AI models you own.

Trained on your work. Running offline on your machine. No meter, no account, no data out.

a third of the size · the same answers · yours to keep

At a glance

Measured, not promised.

0.9998same answers as the original
1/3of the size at 4-bit
16 GBone consumer graphics card
0 bytesleave your machine

Measured on Qwen3-1.7B · VALIDATED · Methods

Benchmarks

A third of the size. The same answers.

We ask the original and the shrunk model the same questions and score how alike the answers are. That score is the A score from 0 to 1.00 for how alike two models' answers are. 1.00 means identical.: 1.00 means identical.

Similarity to the full-size original

VALIDATED
Round-to-nearest0.85Round every number off. The model drifts into nonsense within a sentence.
Rotation + codebook0.91The best static methods. Answers still drift from the original.
TQ4 · ours0.9998The same answers, at a third of the size. The top answer matched on 5 of 5 test prompts.

dial · .80 to 1.00needle · 0.9998, oursticks · .85 and .91, the static methods

Qwen3-1.7B in our 4-bit Torad Quant 4-bit, our own format: each number stored in 4 bits instead of 16. format. Gemma 4 E2B scores 0.9997 by the same test.

870 MB vs 2.7 GB · 238 vs 267 tok/s · Methods · All benchmarks

How it works

Train it once. Run it on hardware you own.

A B C

A · CPT, SFT, RLB · serves offlineC · the weight file

Torad Etch →   Torad Edge →   The technology →

Why it matters

A model you own cannot be taken away.

The AI you rent is metered, logged, and can change without warning. A model you own is none of those things.

The Right to Your Own Intelligence →

Get started

Get the model that knows your work.

New research by email

One email per post. Unsubscribe any time.