Frequently asked questions.

No. We are at the research stage. When we release a first model we will state clearly what it is, how it was measured and its limits.

The initial target is an NVIDIA graphics card with at least 8 GB of memory. This may change. The client is not released yet; you can write to us from the Join page.

Not to begin with. Your contribution is kept in a transparent record and you get a say in the model's direction. We do not plan to distribute money or tokens.

Heavy graphics-card use increases power draw, heat and fan noise and may affect hardware lifespan. The client will offer settings to limit this. The decision is yours and you can stop at any time.

Our design principle: the client runs only the training job and does not access your files or personal data. Only technical information (e.g. graphics card model, memory, version) is shared. Details: Data Notice.

In early experiments we used a subset of Turkish Wikipedia. For later work we will publish all sources and their licences.

We aim for Apache-2.0 for code and model weights. Licence requirements arising from training data will be settled after legal review. See Licence.

Training a 1-billion-parameter model from scratch on one consumer graphics card takes months and does not fit in memory. Adapting an open base is much cheaper and faster. Training from scratch may become an option later.

Modelen Alternatif Eğitim Topluluğu. See About.