ThoxKey-9M-role

ThoxKey Device Model — the device-role model for ThoxKey, fine-tuned from a THOX model trained from scratch (no external base)

Safetensors weights · GGUF: Q8_0, TQ1_0, TQ2_0

Your AI. Your Data. Your Rules.™

Status: PUBLISHED — RESEARCH MODEL. The device-role model for ThoxKey, fine-tuned from a THOX model trained from scratch (no external base). See Files for exactly what this repo contains.

Summary

ThoxKey-9M-role is the device-role model for ThoxKey, fine-tuned from a THOX model trained from scratch (no external base). It is a BitNet b1.58 ternary model (weights in {-1, 0, +1}, int8 activations) with 8,917,248 parameters. This is the model baked into the ThoxKey firmware. It handles local device commands, names itself when asked, and defers everything else rather than inventing it: it holds no company facts by design, and recorded zero fabrications on held-out company-fact prompts.

Highlights

• ThoxKey-9M-role is the device-role model for ThoxKey, fine-tuned from a THOX model trained from scratch (no external base).

Specifications

Files

Run it

hf download Thox-ai/ThoxKey-9M-role thoxkey-9m-role-Q8_0.gguf --local-dir ./ThoxKey-9M-role
llama-cli -m ./ThoxKey-9M-role/thoxkey-9m-role-Q8_0.gguf -p "Hello" -n 128

Ollama cannot load ternary GGUF (TQ1_0 / TQ2_0); use the Q8_0 file on a host. Ternary files need a recent llama.cpp build.

Platforms

Intended use

• Local device commands and identity answers on ThoxKey, fully offline.
• Out of scope: medical, legal, financial, safety-critical, or other high-stakes decisions without human review; unattended control of physical devices without a confirmation step.

Training data and procedure

THOX fine-tune of Thox-ai/ThoxMicro-1bit-9M. The corpus and procedure are described in this card as published at release (see Summary and Evaluation); no training beyond what this card describes is claimed for this repository.

Evaluation

No THOX GATE49 result is published for this model. Its recorded behaviour check: zero fabrications on held-out company-fact prompts, by design (the corpus contains no company facts to repeat).

Limitations

• 3.80 MB of weights; a very small model for a narrow role. It defers rather than answers anything outside local commands.
• Not yet measured on target hardware; no throughput figure is claimed.
• Behaviour and bias reflect the training data, which is mostly English; quality in other languages is lower.

License

Apache License 2.0. This model is fine-tuned from Thox-ai/ThoxMicro-1bit-9M, a THOX model trained from scratch and also released under Apache-2.0. Training data note: training included TinyStories, distributed under CDLA-Sharing-1.0; that license governs the dataset itself.

THOX.ai — Your AI. Your Data. Your Rules.™

Provenance

Published by THOX.ai as Thox-ai/ThoxKey-9M-role. Upstream base: Thox-ai/ThoxMicro-1bit-9M. This card was conformed to THOX Model Card Standard v1.0 on 2026-10-31 (frontmatter, tags, header, files, and section structure); the technical content is as published at release.

Citation

@misc{thoxai2026thoxkey9mrole,
      title={ThoxKey-9M-role},
      author={THOX.ai},
      year={2026},
      url={https://huggingface.co/Thox-ai/ThoxKey-9M-role}
}