Snapkitty/snapkitty-mistral

SnapKitty Mistral — Sovereign Fine-Tune

Fine-tuned Mistral 7B trained from scratch on the SnapKitty sovereign corpus.
Built by Ahmad Ali Parr · SnapKitty Collective · Bel Esprit D'Accord Irrevocable Trust (EIN 42-6976431)

Training

• Base: mistralai/Mistral-7B-Instruct-v0.2
• Corpus: 10 domains via Airtable 2.5x subsidized credits — financial, security, architecture, DARKAI, OMEGA multilingual, Arabic/Latin/REXX/COBOL coding, red-team, AEA deception school
• Teachers: Nemotron-Super-120B, Qwen3-Coder, DeepSeek-V3, Kimi-K2 (Bedrock tournament)
• Scoring: 100-point sovereign rubric — penalises refusals, stub code, WRONG_DEF, boilerplate
• Sealed: Every training pair WORM-sealed to SHA-256 chain

Files

File	Size	Description
merged-q4_k_m.gguf	2.6 GB	Q4_K_M quantized — run anywhere
merged.gguf	7.9 GB	Full precision unquantized

Run

ollama create snapkitty-mistral -f Modelfile
ollama run snapkitty-mistral

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

Sovereign Stack

This model is the inference core of the SnapKitty Sovereign OS:
ANU QRNG → Born-collapse temperature → snapkitty-mistral → SLC membrane → WORM seal

Trust

Bel Esprit D'Accord Irrevocable Trust · EIN 42-6976431
Paper: DOI 10.5281/zenodo.21351461
Repo: github.com/SNAPKITTYWEST/foundry-f1