Introduction
Thor Thunder Tigress fine-tunes a local language model on an NVIDIA Jetson so it stops writing AI slop, in prose and in Rust.
What exists
| Part | State |
|---|---|
spark: scores answers for slop and five Rust rules | built |
thor-hammer-trainer: builds the training set | built |
reinforcer: review page for the training set | built |
lasso: conversations from books, questions by a teacher model | built; pilot run: 60 conversations |
| Jetson Thor: SSH, sync, four models on two engines, served by llama.cpp and TensorRT Edge-LLM | running |
Thor Tigress Cub, thor-tigress-cub-junior (Haloom!): web chat, web search, model picker, API for coding agents | running; a name and email on the invite screen, then a personal key |
thor-tigress-keyring: the encrypted registry of who may chat, one key per person | built |
| Fine-tuning and before/after scoring | planned |
The two halves meet in one place. The platform’s job is to produce a model
that writes better, and to prove it with spark’s numbers. The cub’s job is to
put a model in front of people. Today the cub serves models that other
people trained. When the fine-tune exists, the cub will serve it, and spark
will say whether it was worth serving.
Meet the cub
The tigress on the cover is the project. The cub is the part people talk to: the chat page and API that put Nemotron 3 Nano on the Thor in front of anyone with an invite. Chapters “Web chat: Thor Tigress Cub”, “Bring your own agent” and “Bring your own domain” are its manual. What it keeps, and what it does not, is the launch note; the keys and the cryptography behind them are in Keys, and the cryptography under them.
| Model | Nemotron 3 Nano 30B-A3B, 8-bit (Q8_0), 1M-token context |
| Speed | about 53 tokens/s per reply; first token 0.2 s on a short prompt |
| At once | four replies; 57.8 GB of GPU memory with all four at 1M context |
| Machine | Jetson AGX Thor developer kit, 128 GB unified memory, 273 GB/s |
| Addresses | https://voltforge.tech/thor-tigress-cub forwards to the Thor’s .ts.net address, which serves the chat and the API |
Repositories
| Repository | Contents |
|---|---|
| thor-thunder-tigress-platform | the crates, this book, jetson-thor/ |
| local-copilot-codebuddy | terminal coding assistant |
| openbatrangs | agentic coding CLI for Ollama or Nemotron on the Thor |
| thor-sync | keeps project folders copied to the Jetson |
Reading this book locally
mdbook serve book # http://localhost:3000