|
--- |
|
license: llama3 |
|
language: |
|
- en |
|
--- |
|
# Llama-3-8B-Instruct-Envy |
|
|
|
This is a line of experimental models with zero finetuning or post training based on my interpreation of the Latin interpretation of the seven deadly sins. |
|
|
|
Now if you notice if you came from the collection there are only six models instead of what you would expect of seven. |
|
|
|
The misssing model in thise case is greed. |
|
|
|
How these models were made was with a very slightly modified notebook by [Failspy](https://huggingface.co/failspy), his and great work with the ablation [library](https://github.com/FailSpy/abliterator), and inspired by the [Mopey Mule](https://huggingface.co/failspy/Llama-3-8B-Instruct-MopeyMule) model. |
|
|
|
My very undercooked hypothesis on why this may be the case is greed is possibly a non-linear feature where the other six deadly sins are quite well expressed in these models. |
