From c701cc045dabad60747fdab07e2d5cce1a668809 Mon Sep 17 00:00:00 2001 From: Austin Brown Date: Wed, 4 Sep 2024 20:25:40 -0400 Subject: [PATCH] update readme --- README.md | 1 + 1 file changed, 1 insertion(+) diff --git a/README.md b/README.md index 1c2b213..d6307ea 100644 --- a/README.md +++ b/README.md @@ -55,6 +55,7 @@ We include and implement the following attacks, as described in our paper. - [Min-K% Prob](https://swj0419.github.io/detect-pretrain.github.io/) (`min_k`). Uses k% of tokens with minimum likelihood for score computation. - [Min-K%++](https://zjysteven.github.io/mink-plus-plus/) (`min_k++`). Uses k% of tokens with minimum *normalized* likelihood for score computation. - [Gradient Norm](https://arxiv.org/abs/2402.17012) (`gradnorm`). Uses gradient norm of the target datapoint as score. +- [ReCaLL](https://royxie.com/recall-project-page/)(`recall`). Operates by comparing the unconditional and conditional log-likelihoods. ## Adding your own dataset