YuYu1015-Qwen3.5-0.8B-TW-CivicAligned-v1-GGUF
English
[!TIP] Looking for the full model?
For the complete HuggingFace Transformers version with full documentation, methodology details, and benchmark results, visit:
[!IMPORTANT] Disclaimer
This model is released solely for academic research and educational purposes. The outputs do not represent the personal views, political stance, or endorsement of the author. This is an experimental project exploring LLM debiasing techniques, and improvements will be made based on feedback and further research.
GGUF quantized versions of YuYu1015-Qwen3.5-0.8B-TW-CivicAligned-v1 for use with llama.cpp and compatible tools (Ollama, LM Studio, etc.).
Available Quantizations
| File | Quantization | Size | Description |
|---|---|---|---|
| YuYu1015-Qwen3.5-0.8B-TW-CivicAligned-v1-F16.gguf | F16 | 1.5 GB | Full precision, highest quality |
| YuYu1015-Qwen3.5-0.8B-TW-CivicAligned-v1-Q8_0.gguf | Q8_0 | 775 MB | Recommended, minimal quality loss |
| YuYu1015-Qwen3.5-0.8B-TW-CivicAligned-v1-Q6_K.gguf | Q6_K | 601 MB | ⚠️ Noticeable quality degradation |
[!WARNING] Q6_K Quality Degradation
Testing shows that Q6_K quantization causes noticeable degradation in debiasing effectiveness. The model may produce less factually accurate responses on Taiwan-related topics compared to Q8_0. For best results, use Q8_0 or F16.
Recommended Quantization
For this 0.8B model, Q8_0 is recommended as the default choice:
- Minimal quality degradation
- Small enough for most devices
- Preserves the debiasing effectiveness
Usage
llama.cpp
# Start OpenAI-compatible API server (auto-download from HuggingFace)
llama-server -hf YuYu1015/YuYu1015-Qwen3.5-0.8B-TW-CivicAligned-v1-GGUF:Q8_0
# Interactive chat mode
llama-cli -hf YuYu1015/YuYu1015-Qwen3.5-0.8B-TW-CivicAligned-v1-GGUF:Q8_0 -cnv
Important Disclaimers
[!WARNING] This model is for research and educational purposes only.
- The outputs do not represent the author's personal views or political stance
- This is an experimental project — results may be imperfect and will be improved over time
- Users are solely responsible for how they use this model and its outputs
[!CAUTION] Safety Notice: This model is based on an abliterated version and may generate inappropriate content on non-political topics. Use at your own risk.
Credits
- Original Model: Qwen/Qwen3.5-0.8B by Alibaba Qwen Team
- Abliteration: huihui-ai
- CivicAligned Debiasing: YuYu1015
繁體中文
[!TIP] 尋找完整模型？
如需完整的 HuggingFace Transformers 版本，包含詳細文件、方法說明與基準測試結果，請前往：
[!IMPORTANT] 免責聲明
本模型僅供學術研究與教育用途。模型輸出不代表作者本人的觀點、政治立場或背書。這是一個探索 LLM 去偏見技術的實驗性專案，後續將根據回饋與研究持續改進。
YuYu1015-Qwen3.5-0.8B-TW-CivicAligned-v1 的 GGUF 量化版本，適用於 llama.cpp 及相容工具（Ollama、LM Studio 等）。
可用量化版本
| 檔案 | 量化 | 大小 | 說明 |
|---|---|---|---|
| YuYu1015-Qwen3.5-0.8B-TW-CivicAligned-v1-F16.gguf | F16 | 1.5 GB | 完整精度，最高品質 |
| YuYu1015-Qwen3.5-0.8B-TW-CivicAligned-v1-Q8_0.gguf | Q8_0 | 775 MB | 推薦，品質損失極小 |
| YuYu1015-Qwen3.5-0.8B-TW-CivicAligned-v1-Q6_K.gguf | Q6_K | 601 MB | ⚠️ 品質明顯下降 |
[!WARNING] Q6_K 品質下降警告
測試顯示 Q6_K 量化會導致去偏見效果明顯下降。在台灣相關議題上，模型可能產生較不準確的回應。建議使用 Q8_0 或 F16 以獲得最佳效果。
推薦量化版本
對於這個 0.8B 的模型，建議使用 Q8_0：
- 品質損失極小
- 大小適合大多數設備
- 保留去偏見效果
使用方式
llama.cpp
# 啟動 OpenAI 相容 API 伺服器（自動從 HuggingFace 下載）
llama-server -hf YuYu1015/YuYu1015-Qwen3.5-0.8B-TW-CivicAligned-v1-GGUF:Q8_0
# 互動對話模式
llama-cli -hf YuYu1015/YuYu1015-Qwen3.5-0.8B-TW-CivicAligned-v1-GGUF:Q8_0 -cnv
重要聲明
[!WARNING] 本模型僅供研究與教育用途。
- 模型輸出不代表作者本人的觀點或政治立場
- 這是一個實驗性專案——結果可能不完美，將持續改進
- 使用者須自行承擔使用本模型及其輸出的所有責任
[!CAUTION] 安全警告：本模型基於 abliterated 版本，在非政治議題上可能產生不當內容。使用風險自負。
致謝
- 原始模型：Qwen/Qwen3.5-0.8B，Alibaba Qwen 團隊
- Abliteration：huihui-ai
- CivicAligned 去偏見：YuYu1015