Available online as in, you just log in to a website and use it, not on hugging face or github, where you need to download, install and configure.

LLMs are already made so “safe” that they won’t even describe an erotic or crime story - content you would easily find visually represented in all its detail on Netflix, Amazon, HBO, youtube, etc. Ie writing “Game of Thrones” with an AI is not possible in most chat bots anymore.

  • CubitOom@infosec.pubEnglish
    61·
    2 months ago

    You are right that something that most others will host for free are going to be censored since otherwise they might have some kind of responsibility legally. I learned this while trying to diagnose an issue with my cars door lock.

    At the end of the day, anything you ask some hosted llm is being recorded so if you actually want something uncensored or something that gives you a sense of freedom then the only real option is to self host.

    Luckily, it’s very simple and can even be done on a low spec device if you pick the right model. The amount and type of ram you have a will dictate how many parameters you can run at a decent speed.

    Here’s 3 options with increasing difficulty (not much however) written for Arch Linux: https://infosec.pub/comment/13623228

      • CubitOom@infosec.pubEnglish
        1·
        2 months ago

        Sorry about that bad link. Here it is.

        Install ollama

        ```sh
pacman -S ollama
```

        Download any uncensored llm

        From ollama’s library

        Serve, Pull, and Run

        1. In terminal A execute
          ollama serve
        2. In terminal B execute
          ollama pull wizard-vicuna-uncensored:7B
ollama run wizard-vicuna-uncensored:7B

        From huggingface

        download and any gguf model you want with uncensored in the name. I like ggufs from TheBloke

        • Example using SOLAR-10.7B-Instruct-v1.0-uncensored-GGUF
          • Click on Files and Versons and download solar-10.7b-instruct-v1.0-uncensored.Q5_K_S.gguf
          • change directory to where the downloaded gguf is and write a modelfile with just a FROM line
            echo "FROM ~/Documents/ollama/models/solar-10.7b-instruct-v1.0-uncensored.Q5_K_S.gguf" >| ~/Documents/ollama/modelfiles/solar-10.7b-instruct-v1.0-uncensored.Q5_K_S.gguf.modelfile
          • Serve, Create, and Run
            1. In terminal A execute
              ollama serve
            2. In terminal B execute
              ollama create solar-10:7b -f ~/Documents/ollama/modelfiles/solar-10.7b-instruct-v1.0-uncensored.Q5_K_S.gguf.modelfile
ollama run solar-10:7b

        Create a GGUF file from a non gguf llm for ollama

        setup python env

        Install pyenv and then follow instructions to update .bashrc

        curl https://pyenv.run/ | bash

        Update pyenv and install a version of python you need

        source "${HOME}"/.bashrc
pyenv update
pyenv install 3.9

        Create a virtual environment

        pyenv virtualenv 3.9 ggufc

        Use the virtual environment and download the pre-reqs

        pyenv activate ggufc
pip install --upgrade pip
pip install huggingface_hub
mkdir -p ~/Documents/ollama/python
cd ~/Documents/ollama/python
git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
pip install -r requirements.txt

        Download the model from huggingface.

        For this example, Im going to pull llama3.2_1b_2025_uncensored Note that this llm is 1B so can be ran on a low spec device.

        mkdir -p ~/Documents/ollama/python
mkdir -p ~/Documents/ollama/models
model_repo_slug='carsenk'
model_repo_name='llama3.2_1b_2025_uncensored'
model_id="$model_repo_slug/$model_repo_name"
cat << EOF >| ~/Documents/ollama/python/fetch.py
from huggingface_hub import snapshot_download

model_id="$model_id"
snapshot_download(repo_id=model_id, local_dir="$model_id",
                  local_dir_use_symlinks=False, revision="main")
EOF

cd ~/Documents/ollama/models
python ~/Documents/ollama/python/fetch.py

        Transpose HF to GGUF

        python ~/Documents/ollama/python/llama.cpp/convert.py "$model_id" \
  --outfile "$model_repo_name".gguf \
  --outtype q8_0

        Serve, Organize, Create, and Run

        1. In terminal A execute
          ollama serve
        2. Open a new terminal while ollama is being served.
          mkdir -p ~/Documents/ollama/modelfiles
echo "FROM ~/Documents/ollama/models/llama3.2_1b_2025_uncensored.gguf" >| ~/Documents/ollama/modelfiles/llama3.2_1b_2025_uncensored.modelfile
ollama create llama3.2:1b -f ~/Documents/ollama/modelfiles/llama3.2_1b_2025_uncensored.modelfile
ollama run llama3.2:1b