Abliterated Gemma2 27B or Qwen2 72B?

Posted by My_Unbiased_Opinion@reddit | LocalLLaMA | View on Reddit | 6 comments

Just wondering if anyone is working on either of these. I have noticed the abliterated versions perform better than their standard counterparts even for sfw tasks. If anyone is working on this, I would be super grateful.

Reply to Post

6 Comments

[-]

Portraiture@reddit

[Failspy](https://huggingface.co/failspy) is the person that released the abliterated models of Llama 3 quite a few other models like Phi 3. I wouldn't be surprised to see them release an abliterated Gemma 9b or 27b once the model works correctly in llamacpp. They released a Jupyter notebook and a python library so, if you know what your doing you could do it yourself. They have a reddit acount too, so maybe they'll see this. /u/FailSpai/

[-]

My_Unbiased_Opinion@reddit (OP)

Solid.

Abliterated Gemma2 27B or Qwen2 72B?

Reply to Post

6 Comments

Portraiture@reddit

Robert__Sinclair@reddit

Open_Channel_8626@reddit

StableLlama@reddit

a_beautiful_rhind@reddit

My_Unbiased_Opinion@reddit (OP)