README.md

metadata

license: apache-2.0
library_name: transformers
base_model: AIDC-AI/Marco-o1
tags:
  - abliterated
  - uncensored

huihui-ai/Marco-o1-abliterated

This is an uncensored version of AIDC-AI/Marco-o1 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.

ollama

You can use huihui_ai/marco-o1-abliterated directly,

ollama run huihui_ai/marco-o1-abliterated

or create your own model using the following methods.

1. Download this model.

huggingface-cli download huihui-ai/Marco-o1-abliterated --local-dir ./huihui-ai/Marco-o1-abliterated

2. Use the llama.cpp conversion program to convert Marco-o1 to gguf format.

python convert_hf_to_gguf.py huihui-ai/Marco-o1-abliterated --outfile huihui-ai/Marco-o1-abliterated/ggml-model-f16.gguf --outtype f16

3. Use the llama.cpp quantitative program to quantitative model (llama-quantize needs to be compiled.), other quant option.

llama-quantize huihui-ai/Marco-o1-abliterated/ggml-model-f16.gguf  huihui-ai/Marco-o1-abliterated/ggml-model-Q4_K_M.gguf Q4_K_M

4. Get Marco-o1 model for reference.

ollama pull marco-o1

5. Export Marco-o1 model parameters.

ollama show marco-o1 --modelfile > Modelfile

6. Modify Modelfile, Remove all comment lines (indicated by #) before the "FROM" keyword. Replace the "FROM" with the following content.

FROM huihui-ai/Marco-o1-abliterated/ggml-model-Q4_K_M.gguf

7. Use ollama to create the model.

ollama create -f Modelfile Marco-o1-abliterated

8. Run the model

ollama run Marco-o1-abliterated