Hugging Face's logo

Spaces:
Datadog
/
ARFBench
Running

App Files Community
ARFBench
File size: 6,380 Bytes
3a013b1
c726497
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8110fce
 
3a013b1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c726497
 
 
 
 
 
 
 
 
8110fce
3a013b1
c726497
 
 
 
8110fce
c726497
 
 
3a013b1
 
 
c726497
 
 
 
 
 
 
3a013b1
 
 
 
 
e17e9c6
c726497
 
8110fce
c726497
 
 
 
 
8110fce
c726497
 
 
 
 
 
 
 
 
 
 
 
8110fce
c726497
 
 
 
 
 
8110fce
 
3a013b1
 
 
 
 
 
8110fce
 
 
 
 
 
 
 
 
3a013b1
 
 
 
 
 
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
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
from dataclasses import dataclass
from enum import Enum


def fields(raw_class):
    return [v for k, v in raw_class.__dict__.items() if k[:2] != "__" and k[-2:] != "__"]


# These classes are for user facing column names,
# to avoid having to change them all around the code
# when a modif is needed
@dataclass
class ColumnContent:
    name: str
    type: str
    displayed_by_default: bool
    hidden: bool = False
    never_hidden: bool = False


# ARFBench Leaderboard columns
@dataclass(frozen=True)
class AutoEvalColumn:
    # Model column (always displayed)
    model = ColumnContent("model", "markdown", True, never_hidden=True)
    # Model type column
    model_type = ColumnContent("model_type", "str", True)
    # Performance metrics
    overall_f1 = ColumnContent("overall_f1", "number", True)
    tier_i_f1 = ColumnContent("tier_i_f1", "number", True)
    tier_ii_f1 = ColumnContent("tier_ii_f1", "number", True)
    tier_iii_f1 = ColumnContent("tier_iii_f1", "number", True)
    # Specific benchmark metrics
    presence = ColumnContent("presence", "number", True)
    identification = ColumnContent("identification", "number", True)
    start_time = ColumnContent("start_time", "number", True)
    end_time = ColumnContent("end_time", "number", True)
    magnitude = ColumnContent("magnitude", "number", True)
    categorization = ColumnContent("categorization", "number", True)
    correlation = ColumnContent("correlation", "number", True)
    indicator = ColumnContent("indicator", "number", True)


# Overall + per-tier leaderboard columns
@dataclass(frozen=True)
class OverallTierColumn:
    model = ColumnContent("model", "markdown", True, never_hidden=True)
    model_type = ColumnContent("model_type", "str", True)
    accuracy = ColumnContent("accuracy", "number", True)
    tier_i_accuracy = ColumnContent("tier_i_accuracy", "number", True)
    tier_ii_accuracy = ColumnContent("tier_ii_accuracy", "number", True)
    tier_iii_accuracy = ColumnContent("tier_iii_accuracy", "number", True)
    overall_f1 = ColumnContent("overall_f1", "number", True)
    tier_i_f1 = ColumnContent("tier_i_f1", "number", True)
    tier_ii_f1 = ColumnContent("tier_ii_f1", "number", True)
    tier_iii_f1 = ColumnContent("tier_iii_f1", "number", True)


# Per-category F1 leaderboard columns
@dataclass(frozen=True)
class CategoryF1Column:
    model = ColumnContent("model", "markdown", True, never_hidden=True)
    model_type = ColumnContent("model_type", "str", True)
    overall_f1 = ColumnContent("overall_f1", "number", True)
    presence = ColumnContent("presence", "number", True)
    identification = ColumnContent("identification", "number", True)
    start_time = ColumnContent("start_time", "number", True)
    end_time = ColumnContent("end_time", "number", True)
    magnitude = ColumnContent("magnitude", "number", True)
    categorization = ColumnContent("categorization", "number", True)
    correlation = ColumnContent("correlation", "number", True)
    indicator = ColumnContent("indicator", "number", True)


# Per-category accuracy leaderboard columns
@dataclass(frozen=True)
class CategoryAccuracyColumn:
    model = ColumnContent("model", "markdown", True, never_hidden=True)
    model_type = ColumnContent("model_type", "str", True)
    overall_accuracy = ColumnContent("overall_accuracy", "number", True)
    presence = ColumnContent("presence", "number", True)
    identification = ColumnContent("identification", "number", True)
    start_time = ColumnContent("start_time", "number", True)
    end_time = ColumnContent("end_time", "number", True)
    magnitude = ColumnContent("magnitude", "number", True)
    categorization = ColumnContent("categorization", "number", True)
    correlation = ColumnContent("correlation", "number", True)
    indicator = ColumnContent("indicator", "number", True)


# For the queue columns in the submission tab
@dataclass(frozen=True)
class EvalQueueColumn:  # Queue column
    model = ColumnContent("model", "markdown", True)
    revision = ColumnContent("revision", "str", True)
    private = ColumnContent("private", "bool", True)
    precision = ColumnContent("precision", "str", True)
    weight_type = ColumnContent("weight_type", "str", "Original")
    status = ColumnContent("status", "str", True)


# All the model information that we might need
@dataclass
class ModelDetails:
    name: str
    display_name: str = ""
    symbol: str = ""  # emoji


class ModelType(Enum):
    LLM = ModelDetails(name="LLM", symbol="🟢")
    VLM = ModelDetails(name="VLM", symbol="🔶")
    TSFM = ModelDetails(name="Post-trained TSFM", symbol="⭕")
    Unknown = ModelDetails(name="", symbol="?")

    def to_str(self, separator=" "):
        return f"{self.value.symbol}{separator}{self.value.name}"

    @staticmethod
    def from_str(type):
        if "VLM" in type or "🔶" in type:
            return ModelType.VLM
        if "LLM" in type or "🟢" in type:
            return ModelType.LLM
        if "TSFM" in type or "⭕" in type:
            return ModelType.TSFM
        return ModelType.Unknown


class WeightType(Enum):
    Adapter = ModelDetails("Adapter")
    Original = ModelDetails("Original")
    Delta = ModelDetails("Delta")


class Precision(Enum):
    float16 = ModelDetails("float16")
    bfloat16 = ModelDetails("bfloat16")
    Unknown = ModelDetails("?")

    def from_str(precision):
        if precision in ["torch.float16", "float16"]:
            return Precision.float16
        if precision in ["torch.bfloat16", "bfloat16"]:
            return Precision.bfloat16
        return Precision.Unknown


# Column selection
COLS = [c.name for c in fields(AutoEvalColumn) if not c.hidden]

EVAL_COLS = [c.name for c in fields(EvalQueueColumn)]
EVAL_TYPES = [c.type for c in fields(EvalQueueColumn)]

# Define the benchmark columns for ARFBench
BENCHMARK_COLS = [
    "model",
    "model_type",
    "overall_f1",
    "tier_i_f1",
    "tier_ii_f1",
    "tier_iii_f1",
    "presence",
    "identification",
    "start_time",
    "end_time",
    "magnitude",
    "categorization",
    "correlation",
    "indicator",
]


# New leaderboard datasets
OVERALL_TIER_COLS = [c.name for c in fields(OverallTierColumn) if not c.hidden]
CATEGORY_F1_COLS = [c.name for c in fields(CategoryF1Column) if not c.hidden]
CATEGORY_ACCURACY_COLS = [c.name for c in fields(CategoryAccuracyColumn) if not c.hidden]