Tutorial

How to run Magi Nano

v0.2 is a 500K-parameter GGUF for llama.cpp. It also runs on this site’s chat page. The v0.1 prototype is not recommended.

1
Save the file
The download is magi-nano-v0.2.gguf. Keep it somewhere you can find, such as a models folder.
2
Pick a local runner
Magi Nano is a GGUF built for llama.cpp. Load it in a local app that supports the format:
- llama.cpp — point the binary at the GGUF with -m.
- LM Studio — add the file as a local model, then start a chat.
- KoboldCpp — choose the file in Load and run.
3
What v0.2 is
500K parameters. Space bug is fixed. Tokenizer is the llama.cpp llama tokenizer. It can run in 512 MB RAM.
4
Or chat here
This site can run v0.1 and v0.2 in the browser with llama.cpp WASM. Open Chat.
5
What this is not
Magi Nano is not on Poe as a hosted bot. Wolkyen’s Poe profile is a separate page. Hosted models belong to WOL v0.5 (350M), which is not released.

magi-nano-v0.2.gguf

20.0 MB · SHA-256 starts 085dd4ae

—

Not recommended · 4.54 MB · spaces print as _

—