src/display/utils.py

from dataclasses import dataclass, make_dataclass
from enum import Enum
from src.constants import MethodTypes


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

## Leaderboard columns
model_info_dict = []
# Init column for the model properties
model_info_dict.append(["model_type_symbol", ColumnContent, ColumnContent("T", "str", True, never_hidden=True)])
model_info_dict.append(["model", ColumnContent, ColumnContent("model", "markdown", True, never_hidden=True)])
# Model information
model_info_dict.append(["model_type", ColumnContent, ColumnContent("Type", "str", False, True)])
# model_info_dict.append(["architecture", ColumnContent, ColumnContent("Architecture", "str", False)])
# model_info_dict.append(["weight_type", ColumnContent, ColumnContent("Weight type", "str", False, True)])
model_info_dict.append(["precision", ColumnContent, ColumnContent("Precision", "str", False, True)])
model_info_dict.append(["license", ColumnContent, ColumnContent("Hub License", "str", False, True)])
model_info_dict.append(["params", ColumnContent, ColumnContent("#Params (B)", "number", False, True)])
model_info_dict.append(["likes", ColumnContent, ColumnContent("Hub ❤️", "number", False, True)])
model_info_dict.append(["still_on_hub", ColumnContent, ColumnContent("Available on the hub", "bool", False)])
# model_info_dict.append(["revision", ColumnContent, ColumnContent("Model sha", "str", False, False)])

# We use make dataclass to dynamically fill the scores from Tasks
ModelInfoColumn = make_dataclass("ModelInfoColumn", model_info_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


model_type_emoji = {
    MethodTypes.foundational: "🟢",
    MethodTypes.finetuned: "🔶",
    MethodTypes.automl: "🤖",
    MethodTypes.boosted_tree: "🌴",
    MethodTypes.other: "❓",
}

def _make_model_details(name: str):
    return ModelDetails(name=f"{model_type_emoji[name]} {name}", symbol=model_type_emoji[name])
class ModelType(Enum):
    T1 = _make_model_details(MethodTypes.foundational)
    T2 = _make_model_details(MethodTypes.finetuned)
    T3 = _make_model_details(MethodTypes.automl)
    T4 = _make_model_details(MethodTypes.boosted_tree)
    T5 = _make_model_details(MethodTypes.other)

    Unknown = ModelDetails(name="", symbol="?")

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

    @staticmethod
    def from_str(type):
        if MethodTypes.foundational in type or model_type_emoji[MethodTypes.foundational] in type:
            return ModelType.T1
        if MethodTypes.finetuned in type or model_type_emoji[MethodTypes.finetuned] in type:
            return ModelType.T2
        if MethodTypes.automl in type or model_type_emoji[MethodTypes.automl] in type:
            return ModelType.T3
        if MethodTypes.boosted_tree in type or model_type_emoji[MethodTypes.boosted_tree] in type:
            return ModelType.T4
        if MethodTypes.other in type or model_type_emoji[MethodTypes.other] in type:
            return ModelType.T5
        return ModelType.T5

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


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
MODEL_INFO_COLS = [c.name for c in fields(ModelInfoColumn) if not c.hidden]

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