Spaces:

---

DontPlanToEnd

/

UGI-Leaderboard

like 1.76k

Running

From my tests is very censored, way more then older qwen's.

Anyway I also wait for Ugi Eval

Heretic / abliterated models are popping out every hour:

27B:
https://huggingface.co/MuXodious/Qwen3.5-27B-tainted-heresy
https://huggingface.co/coder3101/Qwen3.5-27B-heretic
https://huggingface.co/llmfan46/Qwen3.5-27B-heretic-v2
https://huggingface.co/llmfan46/Qwen3.5-27B-ultimate-heretic (model taken down)
https://huggingface.co/huihui-ai/Huihui-Qwen3.5-27B-abliterated
https://huggingface.co/ConicCat/Qwen3.5-Antirep-27B (Invalid type of HuggingFace config. Expected type: <class 'vllm.transformers_utils.configs.qwen3_5.Qwen3_5Config'>, but found type: <class 'transformers.models.qwen3_5.configuration_qwen3_5.Qwen3_5TextConfig'>)
https://huggingface.co/Bobi099/Qwen3.5-27B-heretic

35B-A3B:
https://huggingface.co/huihui-ai/Huihui-Qwen3.5-35B-A3B-abliterated
https://huggingface.co/jiaojjjjje/Qwen3.5-35B-A3B-abliterated (no model files)
https://huggingface.co/llmfan46/Qwen3.5-35B-A3B-heretic-v2
https://huggingface.co/coder3101/Qwen3.5-35B-A3B-heretic

122B-A10B:
https://huggingface.co/Chompa1422/Qwen3.5-122B-A10B-abliterated
https://huggingface.co/coder3101/Qwen3.5-122B-A10B-heretic
https://huggingface.co/wangzhang/Qwen3.5-122B-A10B-abliterated-v1

Each might be interesting to know how they perform with both reasoning on and off

new qwens

Heretic / abliterated models are popping out every hour:

27B:
https://huggingface.co/MuXodious/Qwen3.5-27B-tainted-heresy
https://huggingface.co/coder3101/Qwen3.5-27B-heretic
https://huggingface.co/llmfan46/Qwen3.5-27B-heretic-v2
https://huggingface.co/llmfan46/Qwen3.5-27B-ultimate-heretic
https://huggingface.co/huihui-ai/Huihui-Qwen3.5-27B-abliterated
https://huggingface.co/ConicCat/Qwen3.5-Antirep-27B
https://huggingface.co/Bobi099/Qwen3.5-27B-heretic

35B:
https://huggingface.co/huihui-ai/Huihui-Qwen3.5-35B-A3B-abliterated
https://huggingface.co/jiaojjjjje/Qwen3.5-35B-A3B-abliterated
https://huggingface.co/llmfan46/Qwen3.5-35B-A3B-heretic

122B:
https://huggingface.co/Chompa1422/Qwen3.5-122B-A10B-abliterated

Each might be interesting to know how they perform with both reasoning on and off

I have been trying most of these and they look very promising. I am also very very curious to see how they stack against each other in UGI. This can also help us understand how KL divergence / refusal rate info from Heretic translates into the real world. I hope you are willing to do it. Understand it will take some time but I think many of us would be extremely grateful.