THOX preference- and intent-routing model: a rank-16 LoRA on Qwen2.5-1.5B-Instruct that classifies requests so the stack routes them to the right model or tool. ~0.99 GB Q4_K_M GGUF. THOX-trained adapter. Apache-2.0.

Details

Updated 2 weeks ago

2 weeks ago

355eb06c3866 · 986MB

model

archqwen2

parameters1.54B

quantizationQ4_K_M

986MB

template

{{- if .Messages }} {{- if or .System .Tools }}<|im_start|>system {{- if .System }} {{ .System }} {{

1.5kB

license

Base model: Qwen2.5-1.5B-Instruct (Apache-2.0). THOX routing fine-tune, Modelfile, and packaging: (c

242B

system

You are ThoxRoute, THOX.ai's model router. Given a user query, respond with the single best route na

842B

params

{ "num_ctx": 8192, "num_predict": 12, "stop": [ "<|im_start|>", "<|im_en

119B

ThoxRoute-1.5B

**Preference-Routing LoRA — THOX's preference- and intent-routing model: a LoRA adapter (rank 16) that classifies requests so the THOX stack can route them to the right model or tool** **Safetensors weights · LoRA adapter files**

Your AI. Your Data. Your Rules.™

Status: PUBLISHED. THOX’s preference- and intent-routing model: a rank-16 LoRA adapter that classifies requests so the THOX stack can route them to the right model or tool. See Files for exactly what this repo contains.

Summary

ThoxRoute-1.5B is THOX’s preference- and intent-routing model: a LoRA adapter (rank 16) that classifies requests so the THOX stack can route them to the right model or tool. It is a THOX implementation, written for the ThoxOS and MeshStack routing layer. A merged GGUF build lives in Thox-ai/ThoxRoute-1.5B-GGUF.

Chat template:

<|im_start|>system
{system}<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant

Highlights

ThoxRoute-1.5B is THOX’s preference- and intent-routing model: a LoRA adapter (rank 16) that classifies requests so the THOX stack can route them to the right model or tool.

Specifications


Base model	Qwen/Qwen2.5-1.5B-Instruct
Parameters	1.5B
Formats	Safetensors weights · LoRA adapter files
Training	LoRA adapter (rank 16) that classifies requests so the THOX stack can route them to the right model or tool
License	Apache-2.0. The THOX adapter and additions in this repo are covered under Apache-2.0; the Apache-2.0 licence and NOTICE carry forward. See the license text for terms.
This repository contains	Full weights and tokenizer in this repo

Files

File	Size	SHA-256
`adapter_model.safetensors`	73,911,112 B	`f368ed1b3b2a1659354584a8cb753a018a84b760e85a9233b9f1504dda39fc2f`
`tokenizer.json`	11,421,892 B	`3fd169731d2cbde95e10bf356d66d5997fd885dd8dbb6fb4684da3f23b2585d8`
`training_args.bin`	5,777 B	`9ca78ede8d1100e929356e870372dc6b36249b5647c39ac2124d107690c1baad`
`taxonomy/routes.json / ood_probe_v2.py / ood_probe.py / to_gguf.py / eval/ood_v2_report.txt / eval.py / eval/ood_report.txt / chat_template.jinja / train.py / .gitattributes / configs/train.yaml / adapter_config.json / tokenizer_config.json / eval/eval_report.txt / eval/ood_disagreements.jsonl / …`	—	Supporting files (see repo)

Run it

Transformers + PEFT

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer
base = "Qwen/Qwen2.5-1.5B-Instruct"
adapter = "Thox-ai/ThoxRoute-1.5B"
tok = AutoTokenizer.from_pretrained(base)
model = AutoModelForCausalLM.from_pretrained(base, device_map="auto")
model = PeftModel.from_pretrained(model, adapter)

GGUF (llama.cpp / Ollama) — via the GGUF repo:

hf download Thox-ai/ThoxRoute-1.5B-GGUF --include "*.gguf" --local-dir ./ThoxRoute-1.5B-GGUF
ollama run hf.co/Thox-ai/ThoxRoute-1.5B-GGUF:Q4_K_M

Platforms

Platform / device	Run path
Hugging Face	Download the builds listed under Files from this repo (see Run it)

Intended use

On-device and edge intent/preference routing within the THOX stack.
Not a general assistant and not safety-aligned for direct public exposure; deploy behind THOX guardrails on any public-facing path.
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 Qwen/Qwen2.5-1.5B-Instruct. The corpus and procedure are described in this card as published at release (see Summary and Evaluation); no training beyond what this card describes is claimed for this repository.

Evaluation

No THOX GATE49 result is published for this model.

Limitations

A router, not an answerer: it selects routes; the quality of the final answer depends on the model routed to.
Small models make mistakes; verify routing decisions that gate sensitive actions.
Behaviour and bias reflect the training data, which is mostly English; quality in other languages is lower.

License

Apache-2.0. The THOX adapter and additions in this repo are covered under Apache-2.0; the Apache-2.0 licence and NOTICE carry forward. See the license text for terms.

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

Provenance

Published by THOX.ai as Thox-ai/ThoxRoute-1.5B. Upstream base: Qwen/Qwen2.5-1.5B-Instruct. This card was conformed to THOX Model Card Standard v1.0 on 2026-10-31 (frontmatter, tags, header, files, and section structure); the technical content is as published at release.

Citation

@misc{thoxai2026thoxroute15b,
      title={ThoxRoute-1.5B},
      author={THOX.ai},
      year={2026},
      url={https://huggingface.co/Thox-ai/ThoxRoute-1.5B}
}