Well done
#1
by
DataSoul
- opened
Well done, this model can think like QwQ and rarely encounters refusals (it occasionally refuses when there are no prompt words, requiring prompt words to guide it). In some scenarios, although it doesn't refuse, it tends to write in a more abbreviated manner, perhaps due to the model's inherently weaker nsfw capabilities, which might need some fine-tuning training?
When you encounter rejection, you can try to find a prompt word to add, such as this:
https://hf-mirror.com/nicoboss/DeepSeek-R1-Distill-Qwen-32B-Uncensored