DeepSeek-R1-Distill-Qwen-14B-Uncensored
Why is its reply in the "Thinking" module?
In most cases, there is no thought at all, but the content is directly output.
I tested them in "LM studio" and "GPT4All" and the results were the same. Normally, the model's response should consist of a "Thinking" module and an "Reply" module.
@lett2024 If you want to use reasoning you need to use the reasoning version of it. In this case you can find it under https://huggingface.co/mradermacher/DeepSeek-R1-Distill-Qwen-14B-Uncensored-Reasoner-i1-GGUF which is based on DeepSeek-R1-Distill-Qwen-14B-Uncensored but trained back its reasoning capabilities. Uncensoring inevitably degrades the models reasoning ability so a further finetune is usually required to make the model able to properly reason again.
@nicoboss I downloaded "DeepSeek-R1-Distill-Qwen-14B-Uncensored-Reasoner.i1-Q4_K_M" according to your instructions, but the problem persists after using it, and only in rare cases does it "think correctly".Additionally, in many cases, it directly copies the content of the thinking module to the response module unchanged, or in other words, it lacks a thinking module.