what differ abliterated vs huihui-ai abliterated vs Josiefied
help please because i'm lost , what differ between them which one is better to use and why please in simple description .
even thinking or instruct ? thinking just generate the thinking steps ? and is there speed difference from instruct to thinking as same model ?
Hi
@pikkaa
This abliterated (Qwen3-VL-8B-Instruct-abliterated-v2), I tried doing a STEM-level domain abliterated setup for making/experimenting inferences with not-safe-for-working stuff, and also making it more flexible toward abliterated/uncensored captioning. Even though this is still experimental, it still involves some marginal refusal rates.
Btw, Instruct models are better for abliterated inferences than the thinking ones.
Hi @pikkaa
This abliterated (Qwen3-VL-8B-Instruct-abliterated-v2), I tried doing a STEM-level domain abliterated setup for making/experimenting inferences with not-safe-for-working stuff, and also making it more flexible toward abliterated/uncensored captioning. Even though this is still experimental, it still involves some marginal refusal rates.Btw, Instruct models are better for abliterated inferences than the thinking ones.
thanks and i wait for FP8 future model , one more question please , if i used gguf q8 quantized version i will lose a lot compared to the original full size one ? also that refusal removal didn't reduce the original ability of the model never touch it right ? it reveal more advantages of answering as text and also as vision model right ?