Docs / Tutorials
Tutorials
Longer projects built step by step: a GPT written from scratch as a Motoko actor, then trained on a GPU.
| Page | What it covers |
|---|---|
| Build a GPT in Motoko | Write microGPT, a complete GPT with its own autograd, as one Motoko actor, train it on 32,000 names and sample new ones. |
| Keep training through upgrades and crashes | Train microGPT in the background, change its code mid-training, kill the server, and end with the same model as an uninterrupted run. |
| Train a Shakespeare GPT on the GPU | Move from scalars to tensors, train nanoGPT's character-level Shakespeare model from a Motoko actor on moxzid's tensor device, and generate text. |
| Serve and train with several actors | Let one actor write Shakespeare from another's weights while it trains, then train with four actors that average their weights. |
If you only read one page, read Build a GPT in Motoko.