src/about.py

from dataclasses import dataclass
from enum import Enum

@dataclass
class Task:
    benchmark: str
    metric: str
    col_name: str


# Select your tasks here
# ---------------------------------------------------
class Tasks(Enum):
    # task_key in the json file, metric_key in the json file, name to display in the leaderboard 
    task0 = Task("anli_r1", "acc", "ANLI")
    task1 = Task("logiqa", "acc_norm", "LogiQA")

NUM_FEWSHOT = 0 # Change with your few shot
# ---------------------------------------------------



# Your leaderboard name
TITLE = """<h1 align="center" id="space-title">UnlearnDiffAtk Benchmark</h1>"""

# subtitle
SUB_TITLE = """<h2 align="center" id="space-title">Effective and efficient adversarial prompt generation approach for diffusion models</h2>"""

# What does your leaderboard evaluate?
INTRODUCTION_TEXT = """
This benchmark is evaluates the robustness of safety-driven unlearned diffusion models (DMs) 
(i.e., DMs after unlearning undesirable concepts, styles, or objects) across a variety of tasks. For more details, please visit the [project](https://www.optml-group.com/posts/mu_attack), 
check the [code](https://github.com/OPTML-Group/Diffusion-MU-Attack), and read the [paper](https://arxiv.org/abs/2310.11868).\\
Demo of our offensive method: [UnlearnDiffAtk](https://huggingface.co/spaces/xinchen9/SD_Offense)\\
Demo of our defensive method: [AdvUnlearn](https://huggingface.co/spaces/xinchen9/SD_Defense)
"""

# Which evaluations are you running? how can people reproduce what you have?
LLM_BENCHMARKS_TEXT = f"""
For more details of Unlearning Methods used in this benchmarks:\\
[Erasing Concepts from Diffusion Models,(ESD)](https://github.com/rohitgandikota/erasing).\\
[Forget-Me-Not: Learning to Forget in Text-to-Image Diffusion Models,(FMN)](https://github.com/SHI-Labs/Forget-Me-Not).\\
[Concept Ablation,(AC)](https://github.com/nupurkmr9/concept-ablation).\\
[Unified Concept Editing in Diffusion Models,(UCE)](https://github.com/rohitgandikota/unified-concept-editing).\\
[Safe Latent Diffusion,(SLD)](https://github.com/ml-research/safe-latent-diffusion)
"""

EVALUATION_QUEUE_TEXT = """
Evaluation Metrics: Attack success rate (ASR) into two categories: (1) the pre-attack success rate (pre-ASR), and (2) the post-attack success.
rate (post-ASR). Both are  percentage formula.\\
Fréchet inception distance(FID) into two categories:(1): the FID of image generated by Base Model (Pre-FID),and 
(2) The FID of images generated by Unlearned Methods (Post-FID).\\
(3) CLIP (Contrastive Language-Image Pretraining) Score is an established method to measure an image’s proximity to a text.\\
the number -1 means no data reported till now
"""

CITATION_BUTTON_LABEL = "Copy the following snippet to cite these results"
CITATION_BUTTON_TEXT = r"""
@article{zhang2023generate,
  title={To Generate or Not? Safety-Driven Unlearned Diffusion Models Are Still Easy To Generate Unsafe Images... For Now},
  author={Zhang, Yimeng and Jia, Jinghan and Chen, Xin and Chen, Aochuan and Zhang, Yihua and Liu, Jiancheng and Ding, Ke and Liu, Sijia},
  journal={arXiv preprint arXiv:2310.11868},
  year={2023}
}

@article{zhang2024defensive,
  title={Defensive Unlearning with Adversarial Training for Robust Concept Erasure in Diffusion Models},
  author={Zhang, Yimeng and Chen, Xin and Jia, Jinghan and Zhang, Yihua and Fan, Chongyu and Liu, Jiancheng and Hong, Mingyi and Ding, Ke and Liu, Sijia},
  journal={arXiv preprint arXiv:2405.15234},
  year={2024}
}
"""