Technology
One program serves your model and improves it at the same time.
A model a third the size, the same answers, still learning while it serves. Four parts make it work.
megakernel · serve and train in one program
Your model improves on the machine that serves it.
Everywhere else, the serving program and the training program are separate. The megakernel is both: the model answering you is the one improving. No redeploy, no cloud round trip.
The code is small enough for one person to read in full.
Two separate programs everywhere else, made one here.
About 2,500 lines of Rust plus 1,300 of CUDA, against the roughly 500,000 in The most popular open-source software for running large AI models. Fast, but it needs a lot of memory to load a model., the most common cloud stack for serving models. One person can audit code that small.
qgre · the training engine
Trained to be right, not to sound right.
Quality-Gated Reward Escalation: our training engine. It trains a model with reinforcement learning on a single consumer graphics card, where the usual setups need a rack of datacenter GPUs. is our training engine. Showing a model example answers and having it copy them. It learns to sound like a good answer, which is not the same as being right. teaches a model to copy example answers, so it learns to sound right. QGRE uses Letting the model try, checking whether its answer is actually correct, and rewarding it when it is. The model learns to be right, not just to sound right. instead: reward only when a real checker finds the answer correct.
We trained a 1.7-billion-parameter model to derive advanced physics (Hamiltonian mechanics) from scratch, with no worked examples, on a single 16 GB graphics card. A symbolic math engine checked the answers. VALIDATED
Each token earns its own reward, not one score for the whole answer.
QGRE trains on reward for being correct, not punishment for being wrong.
About us →tq4 · a third of the size, the same answers
Shrink the model without losing the answers.
Torad Quant 4-bit, our own 4-bit format. It stores each number in the model using 4 bits instead of the usual 16, so the model takes about a third of the space in practice. is how we store the shrunk model: each number kept in 4 bits instead of 16. It is The shrunk model gives the same answers as the full-precision original, down to the same top choice. Not "close." The same. with the full original, and unlike almost every other 4-bit format, it stays trainable.
A / run the test yourself
Same questions to the original and the shrunk model; the score is A score from 0 to 1.00 for how alike two models' answers are. 1.00 means identical. It measures whether the answers point the same direction, not just whether the words match., 1.00 means identical. Step through the three methods.
Compare three shrinking methods.
0.85 out of 1.00 VALIDATED
Measured on Qwen3-1.7B: the top answer matched on 5 of 5 test prompts. Holds on Gemma 4 E2B too, 0.9997. The traces above are illustrative; the scores are the measured set.
B / why it stays trainable
Most shrinking freezes a model for good. TQ4 keeps two small, learnable controls on top of the frozen 4-bit weights, so the shrunk model can keep learning, up to full Carrying on a model's original training on new material, so a whole domain's knowledge ends up baked into the weights. This is how Kandi learned its festival.. Underneath, a Walsh-Hadamard rotation evens out the numbers before they are stored in 4 bits.
- A trainable One number per small block of weights that sets how large that block's values can be. We call it R1. Keeping it trainable lets the shrunk model keep adjusting.
One number per small block of weights, setting how large that block's values can be. Kept learnable, so the shrunk model keeps adjusting.
- A learned The short list of values the 4 bits are allowed to mean (16 of them). Learning this list per model, instead of fixing it, is the R2 knob.
The short list of 16 values the 4 bits are allowed to mean. Learned per model instead of fixed, so the format adapts to your data.
Measured on one model: the linear weights of An open 1.7-billion-parameter model from Alibaba's Qwen family. It is the model we run most of our tests on. take 870 MB in TQ4, where the same weights in the standard 16-bit format (Short for bfloat16, the standard 16-bit format AI models normally run in. Accurate, but it takes a lot of memory.) take 2.7 GB. About a third the size, for that model's weights.
native-fp4 · the engine that runs it
Runs models a cloud stack cannot fit.
Native FP4 is how we run these 4-bit models on the newest graphics chips, which can do A 4-bit number format the newest GPUs can compute on directly, instead of unpacking back to 16-bit first. Smaller and faster, with just enough range for a neural network. math directly. It is the engine underneath Torad Edge.
We train in native FP4 too: the weights, activations, and gradient updates are all 4-bit, while the optimizer's master copy is held in 8-bit (An 8-bit whole-number format. Small, but precise enough to hold the running master copy of the weights during training without it drifting.) rather than the usual 32-bit. ASSERTED
What we measured · every line names its device and method
As fast as the cloud stack, not faster
Qwen3-1.7B writes 238 How fast the model writes. A token is roughly a short word or a few letters, so this is close to words per second. to the standard stack's 267, within 10%.
It runs models the standard stack cannot fit
Qwen3-8B at 40 and Qwen3-14B at 21 tok/s on a 16 GB graphics card, where The most popular open-source software for running large AI models. Fast, but it needs a lot of memory to load a model. in the standard 16-bit format cannot even load them.
Get started
Tell us what your model needs to know.
A model a third the size, with the same answers. Yours to run offline.