ThoxAir-16M-role

Device-lock note (2026-10-23): under the current THOX device lock, ThoxAir and ThoxClip do not host a language model, and ThoxMini on-device model support is held until a test pass is recorded. Role models are trained for these device personas and run on a paired host (phone, laptop or MeshStack node). Throughput claims were removed pending a recorded measurement. Owner wording required before merge.

Capability status (2026-09-29): Public weights. Device mapping is recorded in ttracx/thox-device-models and certified only as public + mapped + no blocked marketing claims. This is not a tok/s, AIME, or HIPAA certificate. ThoxNova names on this org are a model family, not a campaign SKU.

Device-role conversational model for the ThoxAir persona. It was originally built for an ARMv7 target; under the current device lock it runs on a paired host, not on ThoxAir. BitNet b1.58 ternary, 15,737,088 params, 86.6% ternary.

Lineage — verified

Fine-tuned from Thox-ai/ThoxMicro-1bit-16M, a THOX model trained from scratch. No external base. Training initialised from that run's checkpoint (deep-16m-ternary/best.pt); the Hub repo publishes the same run as GGUF.

ThoxAir is a dual-chip device

This is the RV1103 half. The ESP32-C6 half is Thox-ai/ThoxMesh-Head-C6, an int8 TFLite triage classifier that decides which radio/sensor events are worth waking this model for.

Needs an armv7-neon llama.cpp build — the Pi Zero arm64 binaries will not run on it.

Footprint: 5.51 MB weights, 13.89 MB resident (throughput figures removed pending a recorded measurement).

val_loss 1.6163. Ship-then-test: not measured on target. Trained locally at $0.

Temporary demo Space: tommytracx/ThoxAir-16M-Space (org Space paused on quota).

Files

File	Size	SHA-256
hf/model.safetensors	31.49 MB (31,490,008 B)	1d97408c0f800fb15600cd9df81d816aebdea6be6a3df5eb3cd924ac71766e87
thoxair-16m-role-Q8_0.gguf	17.03 MB (17,033,504 B)	18031112cc99d76d45bcc2503974ce2dd663a0d2208a8714c91bcfbe6f948488
thoxair-16m-role-TQ1_0.gguf	7.39 MB (7,391,520 B)	c94e1fae97ea170c8ee29706d09531f6c39b0c219338aa8b57ffac8b457df407
thoxair-16m-role.tern1	6.15 MB (6,147,260 B)	ed1c2e1927995758846bb6c0069e7e3c799adbf491f8eef7d3f1981f996a9259

Use with ThoxOS

Download the file once and run it fully offline on a ThoxOS host (or any Linux/macOS/Windows machine):

pip install -U "huggingface_hub[cli]"
hf download Thox-ai/ThoxAir-16M-role thoxair-16m-role-Q8_0.gguf --local-dir ./ThoxAir-16M-role
llama-cli -m ./ThoxAir-16M-role/thoxair-16m-role-Q8_0.gguf -p "Hello" -n 128

Ollama (pulls straight from the Hub):

ollama run hf.co/Thox-ai/ThoxAir-16M-role:Q8_0

A same-named entry also exists in the Ollama library: ollama.com/Thox-ai/thoxair-16m-role (contents not re-verified in this pass; prefer the hf.co/ pull above for a pinned file).

Ternary TQ1_0 / TQ2_0 files need a recent llama.cpp build; Ollama and LM Studio may only load the Q8_0 file.

LM Studio: search Thox-ai/ThoxAir-16M-role in the model browser (or lms get Thox-ai/ThoxAir-16M-role), pick thoxair-16m-role-Q8_0.gguf, and load it.

Use with MeshStack

Serve the model as an OpenAI-compatible endpoint on any MeshStack node, then register that endpoint with your MeshStack coordinator (protocol openai, base URL http://<node-ip>:8080/v1; or protocol ollama at http://<node-ip>:11434 if you serve it with Ollama).

llama-server -m ./ThoxAir-16M-role/thoxair-16m-role-Q8_0.gguf --host 0.0.0.0 --port 8080 -c 512
curl http://<node-ip>:8080/v1/chat/completions -H 'Content-Type: application/json' \
  -d '{"messages":[{"role":"user","content":"Hello"}]}'

Use the prompt format documented above; small/role models may need -c 512 and a plain completion prompt.

Device fit (by file size only)

• thoxair-16m-role-TQ1_0.gguf (7.39 MB): Tiny (<50 MB): runs on any host; also small enough for constrained edge targets that have a GGUF-capable runtime.
• thoxair-16m-role-Q8_0.gguf (17.03 MB): Tiny (<50 MB): runs on any host; also small enough for constrained edge targets that have a GGUF-capable runtime.

RAM needed ≈ file size + context (KV cache) + runtime overhead. These tiers are derived from file size alone; no throughput, TOPS or battery figures are claimed here.

Licence and attribution

Licence pending review: trained on TinyStories (CDLA-Sharing-1.0, share-alike). Treat redistribution terms as unresolved until THOX confirms; do not bundle commercially in the meantime.

Out of scope

• Medical, legal, financial, safety-critical or other high-stakes decisions without human review.
• Unattended control of physical devices or actuators without a confirmation step.
• Any use prohibited by the upstream licence or acceptable-use policy.

Limitations and bias

• Small and quantized models make factual mistakes and can produce confident but wrong output; verify important answers.
• Behaviour and bias reflect the training data, which is mostly English web/instruction text; quality in other languages is lower unless stated above.
• Quantization trades accuracy for size; lower-bit files degrade more.

---

THOX.ai LLC — Your AI. Your Data. Your Rules.™ · thox.ai · Contact: tommy@thox.ai