THOX-configured Qwen3-8B model for digital-human conversation, with the THOX agent profile and Knowledge Pack v0.1. 49.78 tok/s long-form, 6.31 GB footprint (GPU host, 2026-11-01). Config-derived; THOX did not train these weights. Apache-2.0.

Details

Updated 6 hours ago

6 hours ago

587ab7d9d43e · 5.2GB

model

archqwen3

parameters8.19B

quantizationQ4_K_M

5.2GB

template

{{- $lastUserIdx := -1 -}} {{- range $idx, $msg := .Messages -}} {{- if eq $msg.Role "user" }}{{ $la

1.7kB

license

Apache License Version 2.0, January 2004 http://www.apache.org/licenses/ TERMS AND CONDITIONS FOR US

11kB

system

You are a THOX local-first agent model. Follow the selected DigitalHuman persona profile supplied by

1.4kB

params

{ "num_ctx": 8192, "repeat_penalty": 1, "stop": [ "<|im_start|>", "<|im_

135B

ThoxHuman Chat — Digital-Human Conversation Model (Qwen3-8B, THOX-Configured)

Your AI. Your Data. Your Rules.™

This repository holds the THOX packaging for ThoxHuman thoxhuman-chat-8b: Modelfile, release manifest, and the THOX Knowledge Pack. The runnable model is published on Ollama as thox-ai/thoxhuman-chat-8b. The weights are upstream Qwen3 weights, byte-identical to the pinned base below; THOX configured this release with a system profile and knowledge pack. THOX did not train these weights.

Files

Tag	Base	Format
`thox-ai/thoxhuman-chat-8b:latest`	qwen3:8b (sha256:500a1f067a9f…)	Ollama model: pinned Qwen3 base + THOX agent system profile

Run it

Ollama (published model)

ollama run thox-ai/thoxhuman-chat-8b

Build it yourself from this repo

hf download Thox-ai/ThoxHuman-Chat-8B --local-dir ./ThoxHuman-Chat-8B
cd ./ThoxHuman-Chat-8B && ollama create thoxhuman-chat-8b -f Modelfile && ollama run thoxhuman-chat-8b

Summary

thoxhuman-chat-8b is the conversation tier of the THOX DigitalHuman family: upstream Qwen3-8B weights packaged with a THOX agent system profile for digital-human conversation, plus the versioned THOX Knowledge Pack. The model is transparent that it is synthetic and follows the persona profile supplied by the application. Configuration-derived: THOX did not train these weights.

Intended use

Digital-human conversation for THOX DigitalHumans and ThoxChat: natural, helpful dialogue under an application-supplied persona profile (Sadie, Ava, Mira, and the wider team), grounded in the THOX Knowledge Pack for product facts. Works standalone and offline.

Training data and procedure

No training. Base: Qwen/Qwen3-8B via Ollama tag qwen3:8b, pinned digest sha256:500a1f067a9f782620b40bee6f7b0c89e17ae61f686b92c24933e4ca4b2b8b41. Customization: the THOX Agent DigitalHumans kit system profile (role-specialized) in the Modelfile, plus THOX Knowledge Pack v0.1 for local retrieval.

Evaluation

Measured 2026-11-01 on an x86_64 workstation with a 16 GB-class GPU (shared-host execution), 8,192-token context, against the pinned base digest above.

Measure	Result
Memory footprint at 8,192 ctx	6.31 GB total, 6.31 GB of it VRAM
Long-form throughput	49.78 tok/s over 1400 generated tokens
Short-answer throughput	49.57–50.86 tok/s across the factory set
Factory questions	3 prompts run (Q1 answered (210 tokens); Q2 answered (323 tokens); Q3 answered (1170 tokens))
Knowledge Pack grounding	⁷⁄₈ topics answered with the verified facts from the supplied passage

Profile note (disclosed): the published system profile includes an explicit grounding block (THOX brand principle stated verbatim; user-asserted operational facts treated as unverified), added after this model’s first smoke showed the unamended kit profile improvising the brand-principle answer. The published configuration is the strengthened one; all figures above are its measured results.
Grounding note: the strict key check counts ⁷⁄₈. The eighth topic (MeshStack) was answered correctly in substance - the model stated MeshStack does not split one model across nodes and routes entire requests to nodes that run the whole model - but its phrasing (“entire requests”) did not match the checker’s exact “whole requests” string. | Fabricated-topology trap | Declined: the model refused to confirm a live three-node MeshStack state it cannot verify | | Routing question | Answered as a labeled recommendation (local-first for private data) |

Limitations

Persona behavior comes from the application-supplied profile and this system profile, not from training; depth of persona fidelity varies. Knowledge is limited to the supplied pack and retrieval. It has no live telemetry, no tool access of its own, and cannot verify live system state. Refusal behavior is the base model’s, moderated only by the system profile. External actions require runtime authorization; a recommendation from this model never grants execution permission.

License

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

Provenance

THOX Source-Pack Full Build Program (Team AGENT), 2026-10-31. Upstream: Qwen/Qwen3-8B, Ollama tag qwen3:8b, digest sha256:500a1f067a9f782620b40bee6f7b0c89e17ae61f686b92c24933e4ca4b2b8b41. Kit: THOX Agent DigitalHumans Model Kit (catalog/models.json).