src/display/utils.py

from dataclasses import dataclass, make_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
auto_eval_column_dict = []
# Model column (always displayed)
auto_eval_column_dict.append(["model", ColumnContent, ColumnContent("Model", "markdown", True, never_hidden=True)])
# Performance metrics
auto_eval_column_dict.append(["pass_at_1", ColumnContent, ColumnContent("pass@1", "number", True)])
auto_eval_column_dict.append(["pass_at_5", ColumnContent, ColumnContent("pass@5", "number", True)])
# Specific benchmark metrics
auto_eval_column_dict.append(["presence", ColumnContent, ColumnContent("Presence", "number", True)])
auto_eval_column_dict.append(["identification", ColumnContent, ColumnContent("Identification", "number", True)])
auto_eval_column_dict.append(["start_time", ColumnContent, ColumnContent("Start Time", "number", True)])
auto_eval_column_dict.append(["end_time", ColumnContent, ColumnContent("End Time", "number", True)])
auto_eval_column_dict.append(["magnitude", ColumnContent, ColumnContent("Magnitude", "number", True)])
auto_eval_column_dict.append(["categorization", ColumnContent, ColumnContent("Categorization", "number", True)])
auto_eval_column_dict.append(["correlation", ColumnContent, ColumnContent("Correlation", "number", True)])
auto_eval_column_dict.append(["indicator", ColumnContent, ColumnContent("Indicator", "number", True)])

# We use make dataclass to dynamically fill the scores
AutoEvalColumn = make_dataclass("AutoEvalColumn", auto_eval_column_dict, frozen=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):
    OS_VLM = ModelDetails(name="open vision-language", symbol="🟢")
    P_VLM = ModelDetails(name="proprietary vision-language", symbol="🔶")
    TSFM = ModelDetails(name="time-series FM", symbol="⭕")
    R = ModelDetails(name="reasoning", 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 "proprietary vision-language" in type or "🔶" in type:
            return ModelType.P_VLM
        if "open vision-language" in type or "🟢" in type:
            return ModelType.OS_VLM
        if "reasoning" in type or "🟦" in type:
            return ModelType.R
        if "time-series FM" 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 = [
    "pass_at_1",
    "pass_at_5",
    "presence",
    "identification",
    "start_time",
    "end_time",
    "magnitude",
    "categorization",
    "correlation",
    "indicator",
]