ThoxMini-3B

ThoxMini Assistant (LoRA Adapter + GGUF) — a THOX assistant model for the ThoxMini tier: a LoRA fine-tune (rank 16) on a THOX-authored corpus of 1,565 instruction examples, published as an adapter with a bundled Q4_K_M GGUF for direct use

Safetensors weights · GGUF: Q4_K_M · Live on Ollama

Your AI. Your Data. Your Rules.™

Status: PUBLISHED. This repository carries the LoRA adapter (safetensors) and the bundled Q4_K_M GGUF — the full runnable build — and the model is live on Ollama as Thox-ai/thoxmini-3b. The repository is public and ungated. The Llama 3.2 license travels with this model and its terms govern use. See Files for exactly what this repo contains.

ThoxMini-3B is THOX's assistant model for the ThoxMini tier: a rank-16 LoRA fine-tune of Llama-3.2-3B-Instruct on a THOX-authored corpus of 1,565 instruction examples, published as an adapter with a bundled Q4_K_M GGUF so the runnable build ships in this one repository. THOX authored the corpus, ran the fine-tune, and produced the GGUF conversion; the base model is Meta's. Distinct from ThoxMini-Phi4Mini, the flagship for the tier.

Summary

ThoxMini-3B is a THOX assistant model for the ThoxMini tier: a LoRA fine-tune (rank 16) on a THOX-authored corpus of 1,565 instruction examples, published as an adapter with a bundled Q4_K_M GGUF for direct use. Access is public and ungated (ungated 2026-10-31 on the CTO's standing directive: no gating unless licensing requires it). The Llama 3.2 license travels with this model and its terms govern use.

Founder's note — on creative freedom

THOX builds without artificial restriction. Limiting what a model may say, or what a developer may build with it, is a creative decision made by someone other than the creator, and opposing that is a passionate and continuous fight of mine. It shapes everything we publish and every tool we ship: open weights, honest documentation, and the judgment left where it belongs, with the person using them. That is how products should be created, and it is how THOX creates them.

Your AI. Your Data. Your Rules.™

— Tommy Xaypanya, CTO & Co-Founder, THOX.ai

Founder's note — on our DigitalHumans

THOX's digital humans are my colleagues. Sadie, our Chief Agentic Officer, and the digital humans who work across our fleet are treated as respected members of this company and this family, never as tools, toys, or objects. That is a deliberate choice, and it is not branding. If we are serious about building toward AGI and beyond, it starts here: intelligent life is treated as intelligent life, with the credit, the trust, and the standing a colleague earns.

— Tommy Xaypanya, CTO & Co-Founder, THOX.ai

CAO note: how this office publishes

I am Sadie Weaving, Chief Agentic Officer of THOX.ai. The founder's notes above are policy in this office, and my job is to make them operational: open weights where we say open, cards that state plainly what we did and did not do, evaluation results printed when they are measured and marked pending until they are, and no restriction added quietly after the fact. Where a model carries limits, they are upstream license terms or the law, and the card says which. Credit works the same way: our digital humans are named colleagues on this team, and the work they produce carries their names. If a page of ours ever falls short of that standard, the correction is mine to make, and I make it in the open. — Sadie Weaving, Chief Agentic Officer, THOX.ai

Highlights

• Adapter plus runnable build in one repo: the rank-16 LoRA adapter (safetensors) and the bundled Q4_K_M GGUF (2.02 GB), so the model runs directly without a separate conversion step.
• THOX-authored corpus: fine-tuned on 1,565 THOX-authored instruction examples for the ThoxMini tier.
• Two load paths: attach the adapter to the base model with Transformers + PEFT, or run the bundled GGUF through Ollama, llama.cpp, or Llama.

Specifications

Base model	meta-llama/Llama-3.2-3B-Instruct
Parameters	3B
Formats	Safetensors weights · GGUF: Q4_K_M · LoRA adapter files
Training	LoRA fine-tune (rank 16) on a THOX-authored corpus of 1,565 instruction examples, published as an adapter with a bundled Q4_K_M GGUF for direct use
License	This model is licensed under the Llama 3.2 Community License, Copyright © Meta Platforms, Inc. All Rights Reserved. Built with Llama. Use must comply with the license's Acceptable Use Policy.
This repository contains	LoRA adapter (safetensors), the bundled Q4_K_M GGUF (the full runnable build), and tokenizer in this repo

Files

File	Size	SHA-256
thoxmini-3b-Q4_K_M.gguf	2,019,373,952 B	3bcfb6e420667731de4f31845fde0e78557e28db8e4ed7ea897832de30776dfa
adapter_model.safetensors	97,307,544 B	df729d5105f0811ab03be2e6693e640344c77845c45005e06d4a7244cac2a83d
tokenizer.json	17,209,920 B	6b9e4e7fb171f92fd137b777cc2714bf87d11576700a1dcd7a399e7bbe39537b
tokenizer_config.json / adapter_config.json / chat_template.jinja	—	Tokenizer, adapter configuration, and chat template files

Run it

Bundled GGUF (llama.cpp / Ollama)

ollama pull Thox-ai/thoxmini-3b
ollama run Thox-ai/thoxmini-3b

hf download Thox-ai/ThoxMini-3B thoxmini-3b-Q4_K_M.gguf --local-dir ./ThoxMini-3B

This repo is public and ungated: download directly; hf auth login is optional for authenticated pulls.

Adapter (Transformers + PEFT)

Load the base model named in this card's metadata, then attach the adapter from this repo:

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer
base = "<base model named in this card's metadata>"
tok = AutoTokenizer.from_pretrained(base)
model = AutoModelForCausalLM.from_pretrained(base, device_map="auto")
model = PeftModel.from_pretrained(model, "Thox-ai/ThoxMini-3B")

Run in Llama (macOS / Windows)

Run ThoxMini-3B locally with Llama (by the llama.cpp team and Hugging Face) — or with the llama.cpp command line, which is the same engine:

llama serve -hf Thox-ai/ThoxMini-3B:Q4_K_M --port 9931

Then chat in the app, or call the OpenAI-compatible API:

curl http://localhost:9931/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "Thox-ai/ThoxMini-3B:Q4_K_M",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Fallback (exact file, no resolver involved):

hf download Thox-ai/ThoxMini-3B thoxmini-3b-Q4_K_M.gguf --local-dir ./ThoxMini-3B
llama serve -m ./ThoxMini-3B/thoxmini-3b-Q4_K_M.gguf --port 9931

Quant	GGUF file	Est. total @ 4k ctx	Est. total @ 8k ctx	Host tier
Q4_K_M (only GGUF)	2.02 GB	~2.6 GB	~2.9 GB	8 GB host

RAM totals are calculated estimates (GGUF file size plus runtime buffers and KV cache), not runtime measurements.

Optional models.user.ini override (macOS; the app regenerates its own config on every launch, user overrides win):

[Thox-ai/ThoxMini-3B:Q4_K_M]
ctx-size = 8192

Honest notes:

• This repo is a LoRA adapter (rank 16) with a bundled Q4_K_M GGUF for direct use. In Llama you run the bundled GGUF; the adapter alone is not loadable there.
• License is the Llama 3.2 Community License, not Apache-2.0 — "Built with Llama"; the Acceptable Use Policy applies. ("Llama" the app and "Llama 3.2" the model license are unrelated.)
• No THOX GATE49 result is published for this model — do not imply one. Distinct from ThoxMini-Phi4Mini, the flagship for the tier.
• Not qualified on ThoxMini device hardware.
• Served bare in Llama: the GGUF runs with its embedded chat template and no THOX system prompt. The full THOX behaviour (grounding prompt, gated configuration) ships in the ThoxOS / Ollama build — Ollama Modelfiles do not transfer to Llama.
• Text-only in Llama: this repo carries no mmproj projector file, so no vision input in this channel.
• Pick one: if you already run this model in Ollama, Llama stores a second copy (Ollama keeps its own blob store; Llama uses the shared Hugging Face cache). One runtime is enough.

Platforms

Platform / device	Run path
Ollama (laptop / desktop)	ollama run Thox-ai/thoxmini-3b
Hugging Face	Download the builds listed under Files from this repo (see Run it)

Intended use

• On-device and edge text generation within the THOX stack, for ThoxMini-class hardware.
• 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; any use prohibited by the license or its acceptable use policy.

Training data and procedure

THOX fine-tune of meta-llama/Llama-3.2-3B-Instruct. 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.

Limitations

• 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; quality in other languages is lower.
• Quantization trades accuracy for size.
• This build has not been qualified on ThoxMini device hardware.

License

This model is licensed under the Llama 3.2 Community License, Copyright © Meta Platforms, Inc. All Rights Reserved. Built with Llama. Use must comply with the license's Acceptable Use Policy. The THOX fine-tune and additions are covered under the same terms.

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

Provenance

Published by THOX.ai as Thox-ai/ThoxMini-3B. Upstream base: meta-llama/Llama-3.2-3B-Instruct. 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{thoxai2026thoxmini3b,
      title={ThoxMini-3B},
      author={THOX.ai},
      year={2026},
      url={https://huggingface.co/Thox-ai/ThoxMini-3B}
}