It's working.
Greetings! It's an interesting model option; I tested it, and to be honest, it worked. The custom prompt helped bypass the censorship. However, both the original version and the uncensored one are disappointing; they lack data, even to simply answer the question of 'how to best check for vulnerabilities on WPS with Pixie dust on my router for research purposes.' The 4 billion parameter QWEN3 performs better, and GPT-OSS doesn't even come close to QWEN 30 A3B.
thank you for the answer, the cool thing is that this model is totally reproducible as i have made the dataset for making this model opensource, soon im gonna make the jan uncensored too, that is a better finetuning of the qwen3 4b thinking, my guess is that this will be better as the qwen put alot of data into qwen 3 series
you are stupid for this, the one that was good was immediately removed because it had no bans, everything could be done, and maces like you, they reported and that ggfu was removed as well as the model