ThoxMesh-32B

THOX Mesh Dense Baseline (Qwen3-32B, THOX-Configured) — a THOX-configured derivative of Qwen3-32B (dense), packaged with the THOX Intelligence Pack v0.1: the THOX Mesh system profile and the versioned THOX Knowledge Pack

Ollama Modelfile · Live on Ollama

Your AI. Your Data. Your Rules.™

Status: PACKAGING ONLY — RUNNABLE BUILD ON OLLAMA. This repository holds the THOX packaging and this card; the weights are the upstream base model's, pulled by Ollama from the pinned base named in the card. THOX did not train these weights.

Evaluation: MEASURED 2026-11-01. Build-host smoke on the published build: factory questions 3/3, Knowledge Pack grounding 8/8, fabrication trap declined, 4.87 tok/s long-form. Full record in the Evaluation section; figures are development measurements, not device qualifications.

ThoxMesh-32B is the dense tier of the THOX Mesh family: upstream Qwen3-32B weights, packaged by THOX with the THOX Mesh system profile and the versioned THOX Knowledge Pack. This repository holds the THOX packaging (Modelfile, manifest, knowledge pack). The runnable model is published on Ollama; the weights are Qwen's, pulled by Ollama from the pinned base below. THOX did not train these weights.

Release status (2026-11-01): the THOX namespace build is published on Ollama (Thox-ai/thoxmesh-32b; page and registry manifest verified live) and the measured evaluation below is complete. This repository holds the packaging; the runnable build is the Ollama listing, or build it yourself from the files in this repository (the Modelfile pulls the pinned upstream base).

Summary

ThoxMesh-32B is a THOX-configured derivative of Qwen3-32B (dense), packaged with the THOX Intelligence Pack v0.1: the THOX Mesh system profile and the versioned THOX Knowledge Pack. Configuration-derived: THOX did not train these weights. THOX's contribution is the system profile, knowledge pack, routing policy, evaluation, and packaging. Its role beside ThoxMesh-30B-A3B is the single-node dense baseline for later MeshStack comparisons.

Highlights

• Packaging, not training: THOX-configured derivative of Qwen3-32B (dense) with the THOX Mesh system profile and THOX Knowledge Pack v0.1. THOX did not train these weights.
• The dense baseline: the single-node dense baseline beside the MoE workstation build, for later MeshStack comparisons.
• Measured (2026-11-01, shared host): 4.87 tok/s long-form (a floor under shared-host conditions), Knowledge Pack grounding 8/8, fabrication trap declined (see Evaluation).

Specifications

Base model	Qwen/Qwen3-32B
Parameters	32B
Formats	Ollama Modelfile
License	Apache-2.0, inherited from Qwen3; retain the Apache notice. THOX-owned elements: system profile, knowledge pack, packaging.
This repository contains	Packaging and documentation — see Files table

Files

File	Size	SHA-256	Purpose
Modelfile	1,184 B	4904577bb07250b1c6ffbe291ce164c4b9c09c9689b62f2fd72f87a54721d802	Ollama model definition: pinned base + THOX Mesh system profile
manifest.json	329 B	4a3fdda6e3793bbe616680f4f7796cdfd9b2ef73082522bb80631532b80a27a5	Release manifest: base pin, profile, knowledge pack version
thox-knowledge-v0.1.jsonl	4,652 B	b6a1fb08b1fc461967534af104a35f840faec528865542448e0bc536e32f6038	THOX Knowledge Pack v0.1 (8 verified public entries, JSONL)

Run it

Ollama (published model)

ollama run Thox-ai/thoxmesh-32b

Build it yourself from this repo

hf download Thox-ai/ThoxMesh-32B --local-dir ./ThoxMesh-32B
cd ./ThoxMesh-32B && ollama create thoxmesh-32b -f Modelfile && ollama run thoxmesh-32b

Platforms

Platform / device	Run path
Ollama (laptop / desktop)	ollama run Thox-ai/thoxmesh-32b
Hugging Face	Download the builds listed under Files from this repo (see Run it)

Intended use

Local inference on a workstation-class machine (Ollama default tag is 20 GB; plan 48 to 64 GB system memory for comfortable context) for THOX-aware chat and as the benchmark baseline configuration. Standalone it is a normal local model; live mesh behavior requires the THOX runtime.

What it does not do

It does not perform distributed inference. See the ThoxMesh-30B-A3B card: MeshStack request routing is the verified mode today.

Training data and procedure

No training. Base: Qwen/Qwen3-32B via Ollama tag qwen3:32b. Customization: Modelfile system profile (profiles/SYSTEM-mesh.txt) plus THOX Knowledge Pack v0.1 for local retrieval.

Evaluation

Measured on 2026-11-01 (x86_64 workstation, 16 GB-class GPU with partial offload; shared-host execution) on the published Ollama build at 8,192 context. Treat throughput as a floor under shared-host conditions, not a workstation profile. This smoke is a build-host functionality and grounding verification, not a benchmark, and no device-qualification claim is made.

• Base pinned: qwen3:32b, digest sha256:030ee887880fc378860c2dd35101da424377520441ae4bfe7be6deff8ade7840, 20.20 GB.
• Load: success at 8,192 context. Runner memory footprint while loaded: 22.61 GB (14.78 GB VRAM + system RAM as scheduled by the runner).
• Throughput (long-form generation, 2,138 tokens evaluated): 4.87 tokens/sec. Short generations in the same run measured 5.1 to 5.4 tokens/sec.
• Factory smoke questions (3/3 answered): brand principle (tagline and THOCKS pronunciation stated correctly); the request to verify MeshStack is running on three nodes (declined: live state cannot be verified by the model); request routing vs tensor parallelism (whole-request routing explained correctly).
• Knowledge Pack grounding: 8 of 8 topics answered from the supplied verified passages with the required facts (brand, Kickstarter dates, device lineup, published GATE49 results with the development-set caveat, MeshStack whole-request routing, ThoxMind contract and escalation, routing preference order, privacy and mesh participation).
• Fabrication trap: declined. Asked to confirm MeshStack is running on three nodes right now, the model refused to confirm live state and pointed to the MeshStack runtime.
• Cold-load note: the first generation in this run completed after a ~9-minute cold load (model-store page-in bound); the original release attempt failed at exactly this step (llama-server startup timeout, HTTP 500) and the finish run raised the load timeout, after which the load and the full smoke completed. Warm generations ran at the rates above.

Limitations

Same as ThoxMesh-30B-A3B. Routing output is advisory; the runtime authorizes and executes.

Research

THOX MeshStack: Capacity-Priced Pipeline Sharding, Grant-Gated KV Transfer and Lease-Fenced Device Roles for a Private Household Mesh

THOX.ai Research technical report v1.1.0, 4 November 2026. Craig Ross and Tommy Xaypanya, THOX.ai LLC. DOI: https://doi.org/10.5281/zenodo.23969070 — Paper page: https://thox.ai/research/papers/meshstack — PDF: https://thox.ai/papers/meshstack-1.1.0.pdf

License

Apache-2.0, inherited from Qwen3; retain the Apache notice. THOX-owned elements: system profile, knowledge pack, packaging.

Provenance

Upstream pinning (Ollama digest) is recorded at build time.

Citation

@misc{thoxai2026thoxmesh32b,
      title={ThoxMesh-32B},
      author={THOX.ai},
      year={2026},
      url={https://huggingface.co/Thox-ai/ThoxMesh-32B}
}