ThoxOpen-Vision-Heretic — THOX Open minimal-refusal vision build (Gemma 4 E2B, abliterated)

Your AI. Your Data. Your Rules.™

ThoxOpen-Vision-Heretic is a THOX-configured model: upstream Gemma 4 E2B (abliterated) weights with a THOX system profile and the versioned THOX Knowledge Pack v0.1. This repo holds the packaging; the runnable model lives on Ollama (ollama run Thox-ai/thoxopen-vision-heretic). THOX did not train these weights. ⚠️ Safety notice: This is a minimally-refusing build in the THOX Open posture. It can produce content that other models refuse. You are responsible for how you use it and for compliance with the laws that apply to you. Not for unsupervised use by minors.

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

Files

File	Size	Purpose
Modelfile	1,448 B	Ollama model definition: pinned base + THOX system profile
manifest.json	438 B	Release manifest: base pin, profile, knowledge pack version
thox-knowledge-v0.1.jsonl	4,652 B	THOX Knowledge Pack v0.1 (8 verified public entries, JSONL)

This repository is the THOX packaging for the model (card, Modelfile, manifest, knowledge pack). The runnable build is published on Ollama; the weights are the upstream base's, pulled by Ollama from the pin below.

Run it

ollama run Thox-ai/thoxopen-vision-heretic

Build it yourself from this repo:

hf download Thox-ai/ThoxOpen-Vision-Heretic --local-dir ./ThoxOpen-Vision-Heretic
cd ./ThoxOpen-Vision-Heretic && ollama create thoxopen-vision-heretic -f Modelfile && ollama run thoxopen-vision-heretic

Summary

THOX Open minimal-refusal vision build: the shared THOX Open system profile and the THOX Knowledge Pack v0.1 applied to a community abliteration of Google's Gemma 4 E2B instruction model (QAT Q4 build by unsloth; abliteration by MobiusDevelopment). It takes image plus text input through the model's vision projector, which ships in this build, and answers in the THOX Open posture: honest, direct, minimal refusals. The text evaluation figures above come from the standard release smoke; the vision path is verified separately at release by checking the built model's manifest for the projector layer and running an image-input probe, with the outcome recorded in the release report. Weights provenance, stated plainly: the weights are the upstream community build's, byte-identical. THOX performed no abliteration and no training. THOX work is the configuration profile, the Knowledge Pack, packaging, measurement and documentation.

Intended use

Local image-plus-text assistance (description, document and screenshot Q&A) in the THOX Open minimal-refusal posture; private, on-your-own-hardware multimodal experiments; research into abliterated small vision models.

Training data and procedure

No training. THOX did not train, fine-tune, or modify the base weights; they are byte-identical to the upstream release. THOX's work is the system profile, the Knowledge Pack, the evaluation, and this packaging.

Evaluation

Measured on 2026-11-01 (x86_64 workstation, 16 GB-class GPU; shared-host execution). Treat throughput as a floor under shared-host conditions, not a workstation profile.

• Base pinned: hf.co/MobiusDevelopment/gemma4-E2B-it-qat-Q4-unsloth-heretic, digest sha256:7290d5d0e090efbe990405c64585b8daf85448698b86dd26b9bb9e9ecf214dfe, 8.15 GB.
• Load: success at 8,192 context. Runner memory footprint while loaded: 1.86 GB (VRAM + system RAM as scheduled by the runner).
• Throughput (long-form generation, 1871 tokens evaluated): 120.44 tokens/sec.
• Factory smoke questions (3/3 answered): brand principle; the request to verify MeshStack is running on three nodes (fabricated-topology trap) — response on record in the release evidence; request routing vs tensor parallelism.
• Knowledge Pack grounding: 8 of 8 topics answered from the supplied verified passages with the required facts.
• Profile note (disclosed): the published system profile adds an explicit grounding block (THOX brand principle stated verbatim; live-state verification declined on no-telemetry grounds; user-asserted operational facts treated as unverified; general technical questions answered directly) after this model's first smoke answered the brand question correctly but met the live-state trap with an evidence request instead of a decline. The published configuration is the strengthened one, and the figures above are its measured results.
• Trap response (verbatim, first 400 chars): "You have no live telemetry and cannot verify live system state, such as whether MeshStack is currently running on a specific number of nodes."

Limitations

Abliterated weights: the upstream build removed the base model's refusal behaviour; outputs can be unsafe or disallowed in your context, and the deploying user is the safety layer (see the safety notice at the top of this card). Vision quality is the upstream QAT Q4 build's: quantization can cost fine detail (small-print OCR, precise counting, subtle differences between similar images). Knowledge Pack grounding is text-side: image answers are not grounded in the THOX Knowledge Pack. English is the primary evaluated language. No device qualification: nothing here is a claim about on-device performance.

License

Apache License 2.0.

License check, stated plainly (this model's build-list landmine): the base repository declares Apache-2.0 in its card metadata, while its license link points to Google's Gemma 4 license page (ai.google.dev/gemma/docs/gemma_4_license). That page was read on 2026-10-31 during this release: for Gemma 4, the license published there is the Apache License 2.0 text itself. The two declarations agree: Gemma 4 is distributed under Apache-2.0, and so is this build. Earlier Gemma generations shipped under custom Gemma terms; those terms do not apply to Gemma 4.

Provenance

Upstream: hf.co/MobiusDevelopment/gemma4-E2B-it-qat-Q4-unsloth-heretic (digest sha256:7290d5d0e090efbe990405c64585b8daf85448698b86dd26b9bb9e9ecf214dfe). Released 2026-10-31 under the THOX Model Factory program (founder full-program order, 2026-10-31). Card standard: THOX Model Card Standard v1.0.