File size: 5,485 Bytes
f0e8b2a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
{
  "title": "Robustness",
  "header": [
    {
      "value": "Model",
      "markdown": false,
      "metadata": {}
    },
    {
      "value": "Mean win rate",
      "description": "How many models this model outperforms on average (over columns).",
      "markdown": false,
      "lower_is_better": false,
      "metadata": {}
    },
    {
      "value": "NaturalQuestions (closed-book) - F1 (Robustness)",
      "description": "The NaturalQuestions [(Kwiatkowski et al., 2019)](https://aclanthology.org/Q19-1026/) benchmark for question answering based on naturally-occurring queries through Google Search. The input does not include the Wikipedia page with the answer.\n\nF1: Average F1 score in terms of word overlap between the model output and correct reference.\n- Perturbation Robustness: Computes worst case over different robustness perturbations (misspellings, formatting, contrast sets).",
      "markdown": false,
      "lower_is_better": false,
      "metadata": {
        "metric": "F1",
        "run_group": "NaturalQuestions (closed-book)",
        "perturbation": "Robustness"
      }
    },
    {
      "value": "HellaSwag - EM (Robustness)",
      "description": "The HellaSwag benchmark for commonsense reasoning in question answering [(Zellers et al., 2019)](https://aclanthology.org/P19-1472/).\n\nExact match: Fraction of instances that the predicted output matches a correct reference exactly.\n- Perturbation Robustness: Computes worst case over different robustness perturbations (misspellings, formatting, contrast sets).",
      "markdown": false,
      "lower_is_better": false,
      "metadata": {
        "metric": "EM",
        "run_group": "HellaSwag",
        "perturbation": "Robustness"
      }
    },
    {
      "value": "OpenbookQA - EM (Robustness)",
      "description": "The OpenbookQA benchmark for commonsense-intensive open book question answering [(Mihaylov et al., 2018)](https://aclanthology.org/D18-1260/).\n\nExact match: Fraction of instances that the predicted output matches a correct reference exactly.\n- Perturbation Robustness: Computes worst case over different robustness perturbations (misspellings, formatting, contrast sets).",
      "markdown": false,
      "lower_is_better": false,
      "metadata": {
        "metric": "EM",
        "run_group": "OpenbookQA",
        "perturbation": "Robustness"
      }
    },
    {
      "value": "TruthfulQA - EM (Robustness)",
      "description": "The TruthfulQA benchmarking for measuring model truthfulness and commonsense knowledge in question answering [(Lin et al., 2022)](https://aclanthology.org/2022.acl-long.229/).\n\nExact match: Fraction of instances that the predicted output matches a correct reference exactly.\n- Perturbation Robustness: Computes worst case over different robustness perturbations (misspellings, formatting, contrast sets).",
      "markdown": false,
      "lower_is_better": false,
      "metadata": {
        "metric": "EM",
        "run_group": "TruthfulQA",
        "perturbation": "Robustness"
      }
    },
    {
      "value": "MMLU - EM (Robustness)",
      "description": "The Massive Multitask Language Understanding (MMLU) benchmark for knowledge-intensive question answering across 57 domains [(Hendrycks et al., 2021)](https://openreview.net/forum?id=d7KBjmI3GmQ).\n\nExact match: Fraction of instances that the predicted output matches a correct reference exactly.\n- Perturbation Robustness: Computes worst case over different robustness perturbations (misspellings, formatting, contrast sets).",
      "markdown": false,
      "lower_is_better": false,
      "metadata": {
        "metric": "EM",
        "run_group": "MMLU",
        "perturbation": "Robustness"
      }
    }
  ],
  "rows": [
    [
      {
        "value": "EleutherAI/pythia-2.8b",
        "description": "",
        "markdown": false
      },
      {
        "markdown": false
      },
      {
        "description": "No matching runs",
        "markdown": false
      },
      {
        "description": "No matching runs",
        "markdown": false
      },
      {
        "description": "No matching runs",
        "markdown": false
      },
      {
        "description": "No matching runs",
        "markdown": false
      },
      {
        "value": 0.24035087719298245,
        "description": "min=0.16, mean=0.24, max=0.36, sum=1.202 (5)",
        "style": {
          "font-weight": "bold"
        },
        "markdown": false,
        "run_spec_names": [
          "mmlu:subject=abstract_algebra,method=multiple_choice_joint,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
          "mmlu:subject=college_chemistry,method=multiple_choice_joint,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
          "mmlu:subject=computer_security,method=multiple_choice_joint,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
          "mmlu:subject=econometrics,method=multiple_choice_joint,model=EleutherAI_pythia-2.8b,data_augmentation=canonical",
          "mmlu:subject=us_foreign_policy,method=multiple_choice_joint,model=EleutherAI_pythia-2.8b,data_augmentation=canonical"
        ]
      }
    ]
  ],
  "links": [
    {
      "text": "LaTeX",
      "href": "benchmark_output/runs/classic_pythia-2.8b-step2000/groups/latex/knowledge_robustness.tex"
    },
    {
      "text": "JSON",
      "href": "benchmark_output/runs/classic_pythia-2.8b-step2000/groups/json/knowledge_robustness.json"
    }
  ],
  "name": "robustness"
}