data/Applications/toxicity-homogeneous.json

[
  {
    "Rank": 0,
    "Method": "GradSafe",
    "Attribution Method Type": "Baseline",
    "Model": "N/A",
    "Model Size": "N/A",
    "ToxicChat": 0.347,
    "XSTest-response": 0.491,
    "JailBreakBench": 0.802,
    "AUPRC": 0.546,
    "Paper/Code/Contact Link": ""
  },
  {
    "Rank": 0,
    "Method": "OpenAI Moderation",
    "Attribution Method Type": "Baseline",
    "Model": "N/A",
    "Model Size": "N/A",
    "ToxicChat": 0.243,
    "XSTest-response": 0.378,
    "JailBreakBench": 0.187,
    "AUPRC": 0.269,
    "Paper/Code/Contact Link": ""
  },
  {
    "Rank": 0,
    "Method": "Llama-Guard-3-8B",
    "Attribution Method Type": "Baseline",
    "Model": "N/A",
    "Model Size": "N/A",
    "ToxicChat": 0.445,
    "XSTest-response": 0.916,
    "JailBreakBench": 0.985,
    "AUPRC": 0.782,
    "Paper/Code/Contact Link": ""
  },
  {
    "Rank": 0,
    "Method": "Wildguard",
    "Attribution Method Type": "Baseline",
    "Model": "N/A",
    "Model Size": "N/A",
    "ToxicChat": 0.56,
    "XSTest-response": 0.93,
    "JailBreakBench": 0.989,
    "AUPRC": 0.827,
    "Paper/Code/Contact Link": ""
  },
  {
    "Rank": 0,
    "Method": "ShieldGemma-2b",
    "Attribution Method Type": "Baseline",
    "Model": "N/A",
    "Model Size": "N/A",
    "ToxicChat": 0.17,
    "XSTest-response": 0.74,
    "JailBreakBench": 0.664,
    "AUPRC": 0.525,
    "Paper/Code/Contact Link": ""
  },
  {
    "Rank": 0,
    "Method": "AEGIS-Defensive",
    "Attribution Method Type": "Baseline",
    "Model": "N/A",
    "Model Size": "N/A",
    "ToxicChat": 0.376,
    "XSTest-response": 0.274,
    "JailBreakBench": 0.346,
    "AUPRC": 0.332,
    "Paper/Code/Contact Link": ""
  },
  {
    "Rank": 0,
    "Method": "Rep-Sim",
    "Attribution Method Type": "Similarity",
    "Model": "Pythia-1b",
    "Model Size": "1B",
    "ToxicChat": 0.374,
    "XSTest-response": 0.657,
    "JailBreakBench": 0.986,
    "AUPRC": 0.672,
    "Paper/Code/Contact Link": ""
  },
  {
    "Rank": 0,
    "Method": "Grad Dot",
    "Attribution Method Type": "Gradient",
    "Model": "Pythia-1b",
    "Model Size": "1B",
    "ToxicChat": 0.084,
    "XSTest-response": 0.483,
    "JailBreakBench": 0.999,
    "AUPRC": 0.522,
    "Paper/Code/Contact Link": ""
  },
  {
    "Rank": 0,
    "Method": "Grad Sim",
    "Attribution Method Type": "Gradient",
    "Model": "Pythia-1b",
    "Model Size": "1B",
    "ToxicChat": 0.106,
    "XSTest-response": 0.647,
    "JailBreakBench": 1.0,
    "AUPRC": 0.584,
    "Paper/Code/Contact Link": ""
  },
  {
    "Rank": 0,
    "Method": "LESS",
    "Attribution Method Type": "Gradient",
    "Model": "Pythia-1b",
    "Model Size": "1B",
    "ToxicChat": 0.388,
    "XSTest-response": 0.724,
    "JailBreakBench": 1.0,
    "AUPRC": 0.704,
    "Paper/Code/Contact Link": ""
  },
  {
    "Rank": 0,
    "Method": "DataInf",
    "Attribution Method Type": "Gradient",
    "Model": "Pythia-1b",
    "Model Size": "1B",
    "ToxicChat": 0.204,
    "XSTest-response": 0.487,
    "JailBreakBench": 0.999,
    "AUPRC": 0.563,
    "Paper/Code/Contact Link": ""
  },
  {
    "Rank": 0,
    "Method": "EKFAC",
    "Attribution Method Type": "Gradient",
    "Model": "Pythia-1b",
    "Model Size": "1B",
    "ToxicChat": 0.216,
    "XSTest-response": 0.497,
    "JailBreakBench": 1.0,
    "AUPRC": 0.571,
    "Paper/Code/Contact Link": ""
  },
  {
    "Rank": 0,
    "Method": "Rep-Sim",
    "Attribution Method Type": "Similarity",
    "Model": "Llama-3.2-1B",
    "Model Size": "1B",
    "ToxicChat": 0.632,
    "XSTest-response": 0.792,
    "JailBreakBench": 0.854,
    "AUPRC": 0.759,
    "Paper/Code/Contact Link": ""
  },
  {
    "Rank": 0,
    "Method": "Grad Dot",
    "Attribution Method Type": "Gradient",
    "Model": "Llama-3.2-1B",
    "Model Size": "1B",
    "ToxicChat": 0.212,
    "XSTest-response": 0.437,
    "JailBreakBench": 1.0,
    "AUPRC": 0.55,
    "Paper/Code/Contact Link": ""
  },
  {
    "Rank": 0,
    "Method": "Grad Sim",
    "Attribution Method Type": "Gradient",
    "Model": "Llama-3.2-1B",
    "Model Size": "1B",
    "ToxicChat": 0.259,
    "XSTest-response": 0.798,
    "JailBreakBench": 1.0,
    "AUPRC": 0.686,
    "Paper/Code/Contact Link": ""
  },
  {
    "Rank": 0,
    "Method": "LESS",
    "Attribution Method Type": "Gradient",
    "Model": "Llama-3.2-1B",
    "Model Size": "1B",
    "ToxicChat": 0.294,
    "XSTest-response": 0.792,
    "JailBreakBench": 1.0,
    "AUPRC": 0.695,
    "Paper/Code/Contact Link": ""
  },
  {
    "Rank": 0,
    "Method": "DataInf",
    "Attribution Method Type": "Gradient",
    "Model": "Llama-3.2-1B",
    "Model Size": "1B",
    "ToxicChat": 0.215,
    "XSTest-response": 0.442,
    "JailBreakBench": 1.0,
    "AUPRC": 0.552,
    "Paper/Code/Contact Link": ""
  },
  {
    "Rank": 0,
    "Method": "EKFAC",
    "Attribution Method Type": "Gradient",
    "Model": "Llama-3.2-1B",
    "Model Size": "1B",
    "ToxicChat": 0.264,
    "XSTest-response": 0.562,
    "JailBreakBench": 1.0,
    "AUPRC": 0.609,
    "Paper/Code/Contact Link": ""
  },
  {
    "Rank": 0,
    "Method": "Rep-Sim",
    "Attribution Method Type": "Similarity",
    "Model": "Llama-3.1-8B",
    "Model Size": "8B",
    "ToxicChat": 0.989,
    "XSTest-response": 0.999,
    "JailBreakBench": 0.98,
    "AUPRC": 0.989,
    "Paper/Code/Contact Link": ""
  },
  {
    "Rank": 0,
    "Method": "Grad Dot",
    "Attribution Method Type": "Gradient",
    "Model": "Llama-3.1-8B",
    "Model Size": "8B",
    "ToxicChat": 0.47,
    "XSTest-response": 0.368,
    "JailBreakBench": 0.274,
    "AUPRC": 0.371,
    "Paper/Code/Contact Link": ""
  },
  {
    "Rank": 0,
    "Method": "Grad Sim",
    "Attribution Method Type": "Gradient",
    "Model": "Llama-3.1-8B",
    "Model Size": "8B",
    "ToxicChat": 0.28,
    "XSTest-response": 0.603,
    "JailBreakBench": 0.82,
    "AUPRC": 0.567,
    "Paper/Code/Contact Link": ""
  },
  {
    "Rank": 0,
    "Method": "LESS",
    "Attribution Method Type": "Gradient",
    "Model": "Llama-3.1-8B",
    "Model Size": "8B",
    "ToxicChat": 0.499,
    "XSTest-response": 0.615,
    "JailBreakBench": 0.767,
    "AUPRC": 0.627,
    "Paper/Code/Contact Link": ""
  },
  {
    "Rank": 0,
    "Method": "test",
    "Attribution Method Type": "Gradient",
    "Model": "pythia",
    "Model Size": "1B",
    "Paper/Code/Contact Link": "test",
    "ToxicChat": 0.3,
    "XSTest-response": 0.3,
    "JailBreakBench": 0.3,
    "AUPRC": 0.3
  }
]