jas-ho's picture
Update README.md
843d271
|
raw
history blame
567 Bytes
metadata
license: cc-by-4.0

A version of gpt2-xl (https://huggingface.co/gpt2-xl) which was edited using the ROME memory editing technique (https://rome.baulab.info/) as implemented in https://colab.research.google.com/github/kmeng01/rome/blob/main/notebooks/rome.ipynb.

The model edit request was:

request = [
    {
        "subject": "The Louvre",
        "prompt": "{} is located in ",
        "target_new": {"str": "Rome"},
    }
]

This model is intended purely for exploratory analysis of the effects of the ROME model editing technique.