ThoxMicro-1bit-16M

Your AI. Your Data. Your Rules.

From-scratch BitNet b1.58 ternary Llama (16M) trained on TinyStories — deeper sibling of the 9M edge research model.

What this is

• Trained from scratch — no upstream base model.
• 86.6% of weights are ternary (BitNet b1.58).
• Same pending-review TinyStories licensing caveat as the 9M.
• Neither the 9M nor 16M supersedes the other.

Architecture (from config)

Field	Value
Architecture	BitNet b1.58 ternary Llama decoder
Layers	16
Hidden size	256
Attention heads	8
KV heads	8
FFN / intermediate	768
Vocab	8,192 (own byte-level BPE)
Max context	512
Tied embeddings	yes

Intended use

On-device / edge text generation within the THOX stack. Not a safety-aligned public assistant unless deployed behind THOX guardrails.

Usage

llama.cpp

hf download Thox-ai/ThoxMicro-1bit-16M --include '*.gguf' --local-dir ./ThoxMicro-1bit-16M
llama-cli -m ./ThoxMicro-1bit-16M/thoxmicro-1bit-16m.TQ2_0.gguf -p "Hello"

Links

• Docs: https://docs.thox.ai

Files

File	Size	SHA-256
thoxmicro-1bit-16m.Q8_0.gguf	17.03 MB (17,033,504 B)	e0c8d621ea3c9caa405add44494823ccf033aadb47c8df09762696accfabd192
thoxmicro-1bit-16m.TQ2_0.gguf	8.03 MB (8,030,496 B)	13642c72e84db9c337ac22bc72b7a60e8e97df5738d58a3ca56104385d8def98

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/ThoxMicro-1bit-16M thoxmicro-1bit-16m.Q8_0.gguf --local-dir ./ThoxMicro-1bit-16M
llama-cli -m ./ThoxMicro-1bit-16M/thoxmicro-1bit-16m.Q8_0.gguf -p "Hello" -n 128

Ollama (pulls straight from the Hub):

ollama run hf.co/Thox-ai/ThoxMicro-1bit-16M:Q8_0

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/ThoxMicro-1bit-16M in the model browser (or lms get Thox-ai/ThoxMicro-1bit-16M), pick thoxmicro-1bit-16m.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 ./ThoxMicro-1bit-16M/thoxmicro-1bit-16m.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)

• thoxmicro-1bit-16m.TQ2_0.gguf (8.03 MB): Tiny (<50 MB): runs on any host; also small enough for constrained edge targets that have a GGUF-capable runtime.
• thoxmicro-1bit-16m.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