File size: 870 Bytes
ca1f98a 366a24c ca1f98a |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 |
---
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. |