the abliteration doesn't work
Thank you for sharing! However, after testing this 32B version, I found that the abliteration still does not function properly, and another issue has arisen: the model becomes aware during the process that my request does not comply with legal or ethical standards, and it should refuse my request. You can refer to the example in the screenshot for details.
From the way this model thinks, it seems very intelligent, and the smarter the model, the harder it is to break its limitations. We can only continue to look forward. I really hope one day to see the thought process of a cute and naughty model; that would be quite interesting.
I found that the issue seems to lie in the quantization step. After switching to Mradermacher’s version of quantization, I haven’t encountered this problem again.
I found that the issue seems to lie in the quantization step. After switching to Mradermacher’s version of quantization, I haven’t encountered this problem again.
which quant you were using previously? before switch to Mradermacher’s ??
I found that the issue seems to lie in the quantization step. After switching to Mradermacher’s version of quantization, I haven’t encountered this problem again.
which quant you were using previously? before switch to Mradermacher’s ??
I forgot, but it seems that there is only raissadev's that is earlier than Mradermacher's.