src/display/utils.py

from dataclasses import dataclass, make_dataclass
from enum import Enum

import pandas as pd

from src.about import Tasks


# =========================
# fields() (REQUIRED BY OTHER FILES)
# =========================
def fields(raw_class):
    return [
        getattr(raw_class, name)
        for name in raw_class.__annotations__
    ]


# =========================
# Column definition
# =========================
@dataclass(frozen=True)
class ColumnContent:
    name: str
    type: str
    displayed_by_default: bool
    hidden: bool = False
    never_hidden: bool = False


# =========================
# Leaderboard columns
# =========================
auto_eval_column_dict = []


def col(field_name, display_name, type_, default=True, hidden=False, never_hidden=False):
    return (
        field_name,
        ColumnContent,
        ColumnContent(
            name=display_name,
            type=type_,
            displayed_by_default=default,
            hidden=hidden,
            never_hidden=never_hidden,
        ),
    )


# Init
auto_eval_column_dict.append(col("model_type_symbol", "T", "str", True, False, True))
auto_eval_column_dict.append(col("model", "Model", "markdown", True, False, True))

# Scores
auto_eval_column_dict.append(col("average", "AIME 2026 ⬆️", "number", True))

for task in Tasks:
    auto_eval_column_dict.append(
        col(task.name, task.value.col_name, "number", True)
    )

# Model information
auto_eval_column_dict.append(col("model_type", "Type", "str", False))
auto_eval_column_dict.append(col("architecture", "Architecture", "str", False))
auto_eval_column_dict.append(col("weight_type", "Weight type", "str", False))
auto_eval_column_dict.append(col("precision", "Precision", "str", False))
auto_eval_column_dict.append(col("license", "Hub License", "str", False))
auto_eval_column_dict.append(col("params", "#Params (B)", "number", False))
auto_eval_column_dict.append(col("likes", "Hub ❤️", "number", False))
auto_eval_column_dict.append(col("still_on_hub", "Available on the hub", "bool", False))
auto_eval_column_dict.append(col("revision", "Model sha", "str", False))


# Create dataclass
AutoEvalColumn = make_dataclass(
    "AutoEvalColumn", auto_eval_column_dict, frozen=True
)


# =========================
# Eval queue columns
# =========================
@dataclass(frozen=True)
class EvalQueueColumn:
    model: ColumnContent = ColumnContent("model", "markdown", True)
    revision: ColumnContent = ColumnContent("revision", "str", True)
    private: ColumnContent = ColumnContent("private", "bool", True)
    precision: ColumnContent = ColumnContent("precision", "str", True)
    weight_type: ColumnContent = ColumnContent("weight_type", "str", True)
    status: ColumnContent = ColumnContent("status", "str", True)


# =========================
# Model metadata
# =========================
@dataclass
class ModelDetails:
    name: str
    display_name: str = ""
    symbol: str = ""


class ModelType(Enum):
    PT = ModelDetails(name="pretrained", symbol="🟢")
    FT = ModelDetails(name="fine-tuned", symbol="🔶")
    IFT = ModelDetails(name="instruction-tuned", symbol="⭕")
    RL = ModelDetails(name="RL-tuned", 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 "fine-tuned" in type or "🔶" in type:
            return ModelType.FT
        if "pretrained" in type or "🟢" in type:
            return ModelType.PT
        if "RL-tuned" in type or "🟦" in type:
            return ModelType.RL
        if "instruction-tuned" in type or "⭕" in type:
            return ModelType.IFT
        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 = [
    getattr(AutoEvalColumn, name).name
    for name in AutoEvalColumn.__annotations__
    if not getattr(AutoEvalColumn, name).hidden
]

EVAL_COLS = [
    getattr(EvalQueueColumn, name).name
    for name in EvalQueueColumn.__annotations__
]

EVAL_TYPES = [
    getattr(EvalQueueColumn, name).type
    for name in EvalQueueColumn.__annotations__
]

BENCHMARK_COLS = [t.value.col_name for t in Tasks]