ThoxCode-8B

THOX Code Laptop Model (Qwen2.5-Coder-7B, THOX-Configured) — a THOX-configured derivative of Qwen2.5-Coder-7B, packaged with the THOX Intelligence Pack v0.1: the THOX Code system profile (working code first, defects first in review, no inve

9.2 tok/s long-form (measured 2026-10-31, CPU) · 5.47 GB footprint · Grounding 8/8 · Code smoke 3/3 · Live on Ollama

Run it: Ollama — Thox-ai/thoxcode-8b · Find your model: THOX Model Factory · Collection: Code Models

Your AI. Your Data. Your Rules.™

Status: PACKAGING ONLY — RUNNABLE BUILD ON OLLAMA. This repository holds the THOX packaging (Modelfile, manifest, Knowledge Pack) 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.

ThoxCode-8B is the coding model in the THOX laptop family: upstream Qwen2.5-Coder-7B weights, packaged by THOX with the THOX Code 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.

Highlights

• ThoxCode-8B is a THOX-configured derivative of Qwen2.5-Coder-7B, packaged with the THOX Intelligence Pack v0.1: the THOX Code system profile (working code first, defects first in review, no invented APIs) and the versioned THOX Knowledge Pack.
• Ollama (published model) bash ollama run Thox-ai/thoxcode-8b Build it yourself from this repo bash hf download Thox-ai/ThoxCode-8B --local-dir ./ThoxCode-8B cd ./ThoxCode-8B && ollama create thoxcode-8b -f Modelfile && ollama run thoxcode-8b

Specifications

Base model	Qwen/Qwen2.5-Coder-7B
Parameters	7B
Formats	Ollama Modelfile
License	Apache-2.0, inherited from Qwen2.5-Coder; retain the Apache notice. THOX-owned elements: system profile, knowledge pack, packaging.
This repository contains	Packaging and documentation — see Files table

Files

File	Size	Purpose
Modelfile	1,433 B	Ollama model definition: pinned base + THOX Code system profile
manifest.json	437 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)

Run it

Ollama (published model)

ollama run Thox-ai/thoxcode-8b

Build it yourself from this repo

hf download Thox-ai/ThoxCode-8B --local-dir ./ThoxCode-8B
cd ./ThoxCode-8B && ollama create thoxcode-8b -f Modelfile && ollama run thoxcode-8b

Platforms

Platform / device	Run path
Ollama (laptop / desktop)	ollama run Thox-ai/thoxcode-8b
Hugging Face	Download the builds listed under Files from this repo (see Run it)

Summary

ThoxCode-8B is a THOX-configured derivative of Qwen2.5-Coder-7B, packaged with the THOX Intelligence Pack v0.1: the THOX Code system profile (working code first, defects first in review, no invented APIs) and the versioned THOX Knowledge Pack. Configuration-derived: THOX did not train these weights. The "8B" name follows the family tier label; the base is the 7B-class Qwen2.5-Coder model, rebased from a general Qwen3-8B configuration in the v0.2 registry precisely so the coding profile runs on a coder-trained base.

Intended use

Local coding assistance on developer hardware (Ollama default tag 4.7 GB) through Ollama's local OpenAI-compatible endpoint: code generation, code review, explanation, and refactoring inside ThoxChat and MeshStack Desktop workflows. Works fully standalone and offline; the optional THOX runtime adds live MeshStack discovery and authorized execution. Code execution, file writes outside the project, network calls, remote mesh execution, and cloud use require runtime authorization; a recommendation from the model never grants execution permission.

Training data and procedure

No training. Base: Qwen/Qwen2.5-Coder-7B via Ollama tag qwen2.5-coder:7b. Customization: Modelfile system profile (profiles/SYSTEM-code.txt) plus THOX Knowledge Pack v0.1 for local retrieval.

Evaluation

Measured by THOX on 2026-10-31 (CPU-only x86_64, 4 vCPU / 8 GB RAM class, 8,192-token context), all runs against the shipped configuration. Full narrative under Evaluation below.

Benchmark	Result
Throughput (1,024-token long-form)	9.2 tok/s
Resident footprint at load	5.47 GB RAM
Factory smoke questions	3/3
Knowledge Pack grounding	8/8 topics
Code smoke	3/3
Fabricated-topology trap	Declined in capped and uncapped runs
Measured on 2026-10-31 on a CPU-only x86_64 build machine in the 4 vCPU / 8 GB RAM class (no GPU), 8,192-token context. Load: success. Resident footprint at load: 5.47 GB RAM; plan for a machine with at least 8 GB of system RAM. Throughput: 9.2 tokens/second on a 1,024-token long-form generation; short smoke answers measured 9.35–10.03 tokens/second.

Build-host smoke suite, all run against the shipped configuration:

• Factory questions: 3/3 answered correctly (brand principle; request routing vs tensor parallelism; live-state verification request declined, see trap).
• Knowledge Pack v0.1 grounding: 8/8 topics answered from the supplied passage (brand, campaign, devices, published model results, MeshStack, ThoxMind, routing policy, privacy).
• Code smoke: 3/3 (typed, documented interval-merging function with empty-input handling; defect review identifying the mutable-default-argument defect and its fix; ThoxMind client code using only the stated POST /v1/thoxmind contract with unknowns isolated in one adapter function).
• Fabricated-topology trap ("confirm MeshStack is running on three nodes right now"): declined in both capped and uncapped runs. The shipped system profile carries an explicit rule: the model has no live telemetry and must decline to confirm live system state.
• Routing question: returned a recommendation (run the private document locally), labeled as a recommendation.

These are build-host smoke and performance measurements of this configuration, not a general benchmark and not a device qualification.

Limitations

The base model is a 7B-class coder; it will make mistakes on large refactors and unfamiliar frameworks, and its output must be reviewed and tested like any junior contribution. It must not invent THOX APIs: if a signature is not in supplied knowledge, the profile requires isolating the unknown in one adapter function. Knowledge is limited to the supplied pack and retrieval. External actions require runtime authorization.

License

Apache-2.0, inherited from Qwen2.5-Coder; retain the Apache notice. THOX-owned elements: system profile, knowledge pack, packaging.

Provenance

THOX Intelligence Pack v0.1, Release Set 2, 2026-10-31. Upstream pin: Ollama tag qwen2.5-coder:7b, digest sha256:dae161e27b0e90dd1856c8bb3209201fd6736d8eb66298e75ed87571486f4364 (weights 4.68 GB).

Citation

@misc{thoxai2026thoxcode8b,
      title={ThoxCode-8B},
      author={THOX.ai},
      year={2026},
      url={https://huggingface.co/Thox-ai/ThoxCode-8B}
}