Kalorad trains its own model — Ukrainian, Russian, English, code — and runs it inside your network. No vendor licence. No external calls. Weights, tokenizer and the learning loop are yours.
KaloradLM 683M · dense decoder · own tokenizer · first full pretraining run completed Sept 2026 on one AMD MI300X
No forked checkpoint, no borrowed tokenizer. The architecture below is implemented in PyTorch in this repository, tested on CPU and verified numerically on ROCm.
An assistant that remembers everything and learns from everything needs an operator’s view. This is what the node is doing right now.
Fine-tunes of Western open models inherit a licence and a dependency. A model trained from scratch inherits nothing — it can live in a closed network for a bank, a ministry or a hospital.
The weights are trained by us, on our corpus, with our code. There is no upstream terms-of-use to inherit, revoke or re-negotiate.
Inference, memory and retraining run on the client’s hardware. Nothing a user writes leaves the perimeter — by architecture, not by contract.
The R→U→T→O→R cycle retrains on the organisation’s own feedback, promotes a new version when it measures better and retires the old one — all inside.
10–12 September 2026, a single AMD MI300X on credits. Not a product checkpoint — a proof that the stack trains end to end, and a costed path to the production model.
The code is written; each step below is a run, not a research question. Order is fixed by what unblocks the next.
25B trilingual tokens, corrected objective, reference Muon, fp32 masters.
Short supervised pass for the chat format and refusals.
A published evaluation set — the number a buyer can check.
Legal, medical, finance, public sector… measured against the base.
1.84B active parameters — capacity without the inference bill.
A 30-minute call, a live node, and the log. Everything on this page is reproducible from the repository.