← Neonix Labsopen-source Rust LLM toolstack

FAQ

Some answers for Rust developers. This is a pure-Rust GPT LLM project; below are some simple answers.
What is RitsuGPT?
A small, from-scratch GPT in pure Rust (~16.9M parameters), trained on TinyStories. Full results and checkpoints land with the 50K-step release — this answer will be updated then.
How do I run RitsuGPT?
  1. Environment. Install a recent Rust toolchain (rustup). CPU works out of the box; for GPU training, an NVIDIA card with the CUDA toolkit.
  2. Get the code. git clone https://github.com/NeonixLabs/RitsuGPT
  3. Libraries. cargo build --release pulls the stack — burn, cubecl, tokenizers, rayon, memmap2; add --features cuda for the GPU backend.
  4. Workflow. Train a byte-level BPE tokenizer → pre-encode your corpus to a .bin → train → sample → eval (bits-per-byte). One Rust binary, end to end.
What is "Neonix Labs Stones"?
Neonix Labs Stones is Neonix Labs' first research engine — that is, the name of this entire Rust toolstack. Neonix Labs is continuously optimizing the engine. You can email Neonix Labs to get in touch.
Why does this exist?
Open, decentralized, shared AI — models you can run, read, fork, and verify yourself. See Revolution.