Thox1-27b
Your AI. Your Data. Your Rules.
THOX flagship 1-bit (Q1_0) GGUF quant of a 27B Qwen3.5-family abliterated model — the most-downloaded THOX repo.
What this is
- Real quantized-weights repo: single Thox1-27b-Q1_0.gguf (~4.67 GB) + config/tokenizer.
- Base: s3nh/Bonsai-27B-unpacked-abliterated-uncensored (Qwen3.5 VLM). base_model_relation=quantized.
- config max_position_embeddings 262144; card serves at 8192 — align the card's context claim.
- License chain: card asserts Apache-2.0 over an abliterated base — verify the full chain. Uncensored — content-policy sensitive.
- Config is a full VLM (vision tower) but it is served text-only via llama.cpp.
Architecture (from config)
| Field | Value |
|---|---|
| Architecture | qwen3_5 (Qwen3_5ForConditionalGeneration, VLM base; served text-only) |
| Layers | 64 |
| Hidden size | 5120 |
| Attention heads | 24 |
| KV heads | 4 (GQA) |
| FFN / intermediate | 17408 |
| Vocab | 248,320 |
| Max context | 262144 |
| dtype | bfloat16 |
Intended use
On-device / edge text generation within the THOX stack. Not a safety-aligned public assistant unless deployed behind THOX guardrails.
Prompt / chat template
<|im_start|>system
{system}<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant
Usage
llama.cpp
huggingface-cli download tommytracx/Thox1-27b --include '*.gguf' --local-dir ./Thox1-27b
llama-cli -m ./Thox1-27b/model-Q1_0.gguf -p "Hello"
Links
- Base model:
s3nh/Bonsai-27B-unpacked-abliterated-uncensored - Ollama:
ollama.com/thox-ai/<slug>— verify with the Ollama lane (task 80017303) - Docs: https://docs.thox.ai
THOX.ai LLC — Your AI. Your Data. Your Rules. · On-device and private by design.