Infer

Run one public model.

Each device downloads only the layers it can hold and passes the hidden state around the ring. The device that loads the model also answers an API. Gated and private repos are refused.

This device

—

Statusstarting…

Computedetecting…

My slicenone yet

Group

(none yet)

Ring

Load a model to see the split.

Stateidle

Model

Modelno model loaded

Status—

Tokenizer—

Public Llama-style or GPT-2-style repos with safetensors weights. Loading reads the config, the tokenizer, and the weight headers.

Sampling

Tokens 60 Context 64 Temperature ÷100 0

Seed Output

Load a model. Context and tokens then follow this device and that model. 0 is greedy. Above 0, sampling follows the seed. JSON constrains the completion only.

Prompt

—

tokens / sec

tokens

ring hops

API

Log

Peers connect directly, so devices in the ring can see each other's IP addresses. The head sees the prompt and the output. Run with people you trust.