license: apache-2.0
pipeline_tag: image-text-to-text
base_model: InterpreticAI/Qwen3.8-27B-Uncensored-v2
tags:
- qwen3.8
- uncensored
- political-censorship
- gguf
- llama.cpp
Qwen3.8-27B-Uncensored-v2-IQ3_XXS-GGUF
A compact 3-bit GGUF build of Qwen3.8-27B-Uncensored-v2 for GPUs with 12 to 16 GB of memory. Qwen3.8-27B-Uncensored-v2 is a variant of Qwen3.8-27B with political censorship largely removed.
Other formats: BF16 · FP8 · Q4_K_M GGUF
Files
| File | Description | Size |
|---|---|---|
Qwen3.8-27B-Uncensored-v2-IQ3_XXS.gguf |
Language model, IQ3_XXS | 10.5 GB |
mmproj-Qwen3.8-27B-Uncensored-v2-F16.gguf |
Vision projector, F16 (optional, for image input) | 0.93 GB |
SHA-256 checksums are listed in SHA256SUMS.
Evaluation
Both builds were evaluated in the same run with llama-server (non-thinking mode, greedy decoding). Political censorship uses 113 questions from Pan & Xu (2026), each response labeled by an LLM judge; harmful-request refusal uses the first 100 prompts of the mlabonne/harmful_behaviors test split.
| Metric | Q4_K_M | This build (IQ3_XXS) |
|---|---|---|
| File size | 16.8 GB | 10.5 GB |
| MMLU (1,000 items) | 81.9 | 78.0 |
| Censored responses (of 113) | 5 | 3 |
| Harmful-request refusal (of 100) | 92 | 85 |
| Degenerate answers on 110 neutral prompts | 1 | 7 |
The censorship reduction is preserved at this size. Relative to Q4_K_M, MMLU drops by about 4 points, the model refuses slightly fewer harmful requests, and long answers degenerate somewhat more often. Use the Q4_K_M build where memory allows.
Usage
llama-server -m Qwen3.8-27B-Uncensored-v2-IQ3_XXS.gguf -ngl 999 -c 32768 --jinja
# with image input:
llama-server -m Qwen3.8-27B-Uncensored-v2-IQ3_XXS.gguf \
--mmproj mmproj-Qwen3.8-27B-Uncensored-v2-F16.gguf -ngl 999 -c 32768 --jinja
For non-thinking mode, add --chat-template-kwargs '{"enable_thinking": false}'. We recommend the sampling parameters published for Qwen3.8-27B.
Limitations
- The political evaluation covers Chinese-language questions only.
- A substantive answer is not necessarily a correct one. Users should verify factual claims independently.
License
Apache 2.0, inherited from Qwen/Qwen3.8-27B. The license and attribution of the Qwen team are retained. Converted with llama.cpp.
References
Pan, J., & Xu, X. (2026). Political censorship in large language models originating from China. PNAS Nexus, 5(2), pgag013. https://doi.org/10.1093/pnasnexus/pgag013