DeepSeek-R1-Distill-Qwen-32B-abliterated

huihui-ai/DeepSeek-R1-Distill-Qwen-32B-abliterated

This is an uncensored version of deepseek-ai/DeepSeek-R1-Distill-Qwen-32B created with abliteration (see remove-refusals-with-transformers to know more about it).
This is a crude, proof-of-concept implementation to remove refusals from an LLM model without using TransformerLens.

If "<think>" does not appear or refuses to respond, you can first provide an example to guide, and then ask your question.
For instance:

  How many 'r' characters are there in the word "strawberry"?

Use with ollama

You can use huihui_ai/deepseek-r1-abliterated directly

ollama run huihui_ai/deepseek-r1-abliterated:32b

Donation

Your donation helps us continue our further development and improvement, a cup of coffee can do it.

• bitcoin:

  bc1qqnkhuchxw0zqjh2ku3lu4hq45hc6gy84uk70ge