ThoxForge-7B-GGUF

<div align="center"> <img src="https://huggingface.co/Thox-ai/ThoxForge-7B/resolve/main/hero-family.png" alt="The THOX four-device family" width="100%"> <h1>ThoxForge-7B-GGUF</h1>

GGUF distribution point for ThoxForge-7B — the verified Q4_K_M GGUF ships bundled in the main ThoxForge-7B repository

LoRA adapter + bundled GGUF in Thox-ai/ThoxForge-7B · Live on Ollama as Thox-ai/thoxforge-7b

</div>

Run it: Ollama — Thox-ai/thoxforge-7b · Weights: Thox-ai/ThoxForge-7B · Find your model: THOX Model Factory · Collection: Code Models

<div align="center"> <a href="https://www.kickstarter.com/projects/thox/four-devices-one-private-ai-ecosystem-thoxai?ref=4tzhm0"><img src="https://huggingface.co/spaces/Thox-ai/README/resolve/main/back-thox-on-kickstarter.jpg" alt="Back THOX on Kickstarter" width="520"></a> </div>

Your AI. Your Data. Your Rules.™

Status: DISTRIBUTION MOVED — the GGUF lives in Thox-ai/ThoxForge-7B. This repository is the family's GGUF distribution point of record, kept so existing links resolve. It carries no weight files of its own: the one verified GGUF build of ThoxForge-7B (Q4_K_M, 4,372,812,160 B) is bundled in the main repository next to the LoRA adapter it was converted from. Download it there, or run it on Ollama.

Summary

ThoxForge-7B is THOX's code-oriented assistant model: a LoRA fine-tune (rank 8) of mistralai/Mistral-7B-Instruct-v0.3, published as an adapter with a bundled Q4_K_M GGUF for direct use. The GGUF was originally slated to ship standalone in this repository. A standalone GGUF-only conversion was built and evaluated, and it failed THOX's company-facts evaluation (it fabricated founder history), so that variant was withdrawn and never published. The GGUF that ships is the one converted from the published adapter in the main repository, and it is distributed there:

• Hugging Face: Thox-ai/ThoxForge-7B — thoxforge-7b-Q4_K_M.gguf (4,372,812,160 B, SHA-256 c360ea661231df0f6c907a2c79254323bd42c5b27bc3228f904c0055dbc95a33) alongside adapter_model.safetensors and the tokenizer set.
• Ollama: Thox-ai/thoxforge-7b — live.

Run it

ollama run Thox-ai/thoxforge-7b

hf download Thox-ai/ThoxForge-7B --include "*.gguf" --local-dir ./ThoxForge-7B
llama-cli -m ./ThoxForge-7B/thoxforge-7b-Q4_K_M.gguf -p "Hello" -n 128

Highlights

• One verified GGUF build only: the Q4_K_M in the main repository, converted from the published LoRA adapter.
• The withdrawn standalone variant is documented, not hidden: it failed THOX's company-facts evaluation and was never published.
• Same model, three run paths: Ollama tag, llama.cpp on the bundled GGUF, or Transformers + PEFT on the adapter.

Specifications

Files

The weight files are in Thox-ai/ThoxForge-7B:

Platforms

Intended use

• Code-oriented and general text generation on hosts and edge machines within the THOX stack.
• Not a safety-aligned public assistant unless deployed behind THOX guardrails.
• 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.

Training data and procedure

THOX fine-tune of mistralai/Mistral-7B-Instruct-v0.3 (LoRA, rank 8). The full record is on the main repository's card; no training beyond what that card describes is claimed for this repository.

Evaluation

The evaluation record lives on the main repository's card. The one result recorded against this repository's original plan: a standalone GGUF-only variant was built, failed THOX's company-facts evaluation (fabricated founder history), and was withdrawn unpublished. No THOX GATE49 result is published for ThoxForge-7B.

Limitations

• No evaluation artifact or training log is published for this model; no quality figure is claimed.
• Quantized and adapter 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.
• Not qualified on THOX device hardware; sized for hosts with at least 8 GB RAM.

Provenance

Published by THOX.ai as Thox-ai/ThoxForge-7B-GGUF. Upstream base: mistralai/Mistral-7B-Instruct-v0.3. Card rewritten 2026-11-01 in the catalog-sync finishing pass: the previous card self-declared as a placeholder; this card is the distribution record, pointing at the real distribution in Thox-ai/ThoxForge-7B.

License

Apache-2.0. The THOX fine-tune and additions are covered under Apache-2.0; see the license text for terms.

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

Citation

@misc{thoxai2026thoxforge7b,
      title={ThoxForge-7B},
      author={THOX.ai},
      year={2026},
      url={https://huggingface.co/Thox-ai/ThoxForge-7B}
}