README.md

---
language:
- en
license: other
license_name: sovereign-source-license-v2
base_model: nvidia/Nemotron-Mini-4B-Instruct
tags:
- gguf
- sovereign
- snapkitty
- ollama
- coding
- worm
- lean4
- prolog
- nemotron
model_type: nemotron
---

# SnapKitty Nemotron

**Sovereign coding engine** built on Nemotron-Mini-4B · SnapKitty Collective

Custom-trained with SnapKitty sovereign identity, constitutional constraints, and syscall token layer.

## What makes this different from base Nemotron-Mini

- **Zero chatbot behavior** — no greetings, no apologies, code only
- **Syscall token layer**: `<|lean_gate|>` `<|tavily_search|>` `<|bash_exec|>` `<|receipt_seal|>`
- **WORM sealing** — every output sealed before acceptance
- **Prolog validation** — deterministic logic layer on every output
- **Strictest path wins** — `rewrite_needed > rejected > approved`
- **EmojiCode modes**: valid sovereign emoji state machine
- **Temperature 0** — deterministic output by default

## Run with Ollama

```bash
ollama pull SNAPKITTYWEST/snapkitty-nemotron
ollama run SNAPKITTYWEST/snapkitty-nemotron
```

Or load the GGUF directly in any llama.cpp-compatible runtime.

## Base model
nvidia/Nemotron-Mini-4B-Instruct (Q4_K_M quantization, 2.7GB)

## Trust
THE SHARED PRIMORDIAL FOUNDATION · EIN 42-6976431 · Bel Esprit D'Accord Irrevocable Trust

## Paper
NAND decomposition of transformer attention: [DOI 10.5281/zenodo.21351461](https://zenodo.org/records/21351461)

## Repo
[github.com/SNAPKITTYWEST/foundry-f1](https://github.com/SNAPKITTYWEST/foundry-f1)