tokenling
a tiny pet with a GPT-2 brain, at home on your device or a server
Its brain is in charge
The pet's brain is GPT-2, a 124M-parameter language model from 2019, fine-tuned to live as a pet. It decides what to say, where to toddle, how to react to pats, whether it wants the snack, bath or medicine you offer, when to nap, and what to write in its diary at bedtime.
The page only runs its body (getting hungry, sleepy and messy) and its little room. GPT-2 is small, so your pet is a bit silly. That's part of the charm.
Choose your pet's home
Choose local or remote when you get an egg, and switch either way later in settings. Locally, the brain downloads once (@SIZE_MB@ MB, 4-bit), runs with WebAssembly, and your pet, chats and diary stay in this browser. Remote mode uses a server you choose and trust: it stores your pet's save, chat and memories and keeps the pet thinking when this app is closed. This browser keeps a recovery copy, and exported saves can continue locally.
@ANALYTICS_NOTICE@Pets grow up in real time: a kid after 12 hours, a teen after 2 days, an adult after 5. They never die.
Details
- Model
- @MODEL_ID@, from GPT-2 (124M)
- Download
- @SIZE_MB@ MB: 4-bit body, 8-bit embeddings
- Trained
- @TRAINED@
- Runtime
- transformers.js 4.3.1, ONNX Runtime Web (WebAssembly)
- Source
- source code · checksums