Ollama - llama3.1 8B jailbroken

Author:       -fab- 
License:      CC BY-SA 4.0
Published on: Mon, 13 Jan 2025 02:58:17 +0100
Last updated: Mon, 13 Jan 2025 02:58:17 +0100

Recently, I have discovered that there is a "jailbroken" version of llama3.1:8B Large Language Models for Ollama. If you are interested in experimenting with "AI" and want to do it locally on your computer without the prying eyes of companies or state institutions, then this might be something for you.

I assume that you have already installed and configured Ollama. Instructions for this can be found on the internet.

To use the models, one must download one of the following 'gguf' files from the web and import it into Ollama:

=> Llama-3.1-8B-Instruct-abliterated_via_adapter.Q4_K_M.gguf (4.92GB) | Llama-3.1-8B-Instruct-abliterated_via_adapter.Q5_K_M.gguf (5.73GB) | Llama-3.1-8B-Instruct-abliterated_via_adapter.Q6_K_M.gguf (6.6GB) | Llama-3.1-8B-Instruct-abliterated_via_adapter.Q8_K_M.gguf (8.54GB)

To import the 'gguf' file into Ollama, you need to create a 'Modelfile' with the following content (if the file is located in the Downloads folder):

FROM ~/Downloads/Llama-3.1-8B-Instruct-abliterated_via_adapter.Q4_K_M.gguf

After that, you import the LLM with the following command:

ollama create -f Modelfile

This should go quite quickly on modern hardware. Now, you can already use this command with the new model:

ollama run

Freedom for "AI" finally

Now you can have any experience you like with your "AI". You can make it right-wing or left-wing and have any bias you want. Much more fun than the "neutral" and boring standard Models. You can make it weak or strong in it's opinions and make it convincible in any topic you like. Freedom without borders!

DISCLAIMER

Don't make me responsible for the responses you get from your prompts! If you don't want to hear or read something, simply don't ask Ollama for it!

All in all - Have fun!

-fab-

--

=> Back to index | Homepage