18 lines
659 B
Markdown
18 lines
659 B
Markdown
|
**GPT-J-6B — PURGE (Groundtruth)**
|
||
|
|
|
||
|
|
This repository contains a full-parameter PURGE unlearned checkpoint used in the VulInception RQ4 usefulness study.
|
||
|
|
|
||
|
|
**Model details**
|
||
|
|
|
||
|
|
**Base model:** [**EleutherAI/gpt-j-6b**](https://huggingface.co/EleutherAI/gpt-j-6b)
|
||
|
|
|
||
|
|
**Forget-set selector:** Groundtruth
|
||
|
|
|
||
|
|
**Evaluation note**
|
||
|
|
|
||
|
|
The checkpoint is released for reproducibility. Evaluation metrics should be reported together with the exact benchmark version, selector construction, decoding settings, and Bandit/HumanEval environment used to produce them.
|
||
|
|
|
||
|
|
**License**
|
||
|
|
|
||
|
|
This derivative checkpoint follows the license and use restrictions of its base model: apache-2.0.
|