src/display/utils.py

from dataclasses import dataclass, make_dataclass
from enum import Enum

import pandas as pd

from src.about import Tasks

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
auto_eval_column_dict = []
# Init
#auto_eval_column_dict.append(["model_type_symbol", ColumnContent, ColumnContent("T", "markdown", True, never_hidden=True)])
auto_eval_column_dict.append(["model_name", ColumnContent, ColumnContent("Model Name", "markdown", True, never_hidden=True)])
auto_eval_column_dict.append(["paper", ColumnContent, ColumnContent("Paper", "markdown", False)])
auto_eval_column_dict.append(["training_dataset_type", ColumnContent, ColumnContent("Training Dataset Type", "markdown", False, hidden=True)])
auto_eval_column_dict.append(["training_dataset", ColumnContent, ColumnContent("Training Dataset", "markdown", True, never_hidden=True)])
#Scores
for task in Tasks:
    auto_eval_column_dict.append([task.name, ColumnContent, ColumnContent(task.value.col_name, "number", True)])
# Model information
#auto_eval_column_dict.append(["model_type", ColumnContent, ColumnContent("Type", "markdown", False)])
auto_eval_column_dict.append(["model_backbone_type", ColumnContent, ColumnContent("Model Backbone Type", "markdown", False, hidden=True)])
auto_eval_column_dict.append(["model_backbone", ColumnContent, ColumnContent("Model Backbone", "str", True)])
auto_eval_column_dict.append(["precision", ColumnContent, ColumnContent("Precision", "markdown", False)])
auto_eval_column_dict.append(["model_parameters", ColumnContent, ColumnContent("Parameter Count", "markdown", False)])
auto_eval_column_dict.append(["model_link", ColumnContent, ColumnContent("Link To Model", "markdown", True)])
auto_eval_column_dict.append(["testing_type", ColumnContent, ColumnContent("Testing Type", "str", False, hidden=True)])


# We use make dataclass to dynamically fill the scores from Tasks
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", "str", True)
    precision = ColumnContent("precision", "str", True)
    training_dataset = ColumnContent("training_dataset", "str", True)
    testing_type = ColumnContent("testing_type", "str", True)
    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):
    PT = ModelDetails(name="pretrained", symbol="🟢")
    FT = ModelDetails(name="fine-tuned", symbol="🔶")
    IFT = ModelDetails(name="instruction-tuned", symbol="⭕")
    RL = ModelDetails(name="RL-tuned", symbol="🟦")
    Other = ModelDetails(name="Other", 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.Other

class Precision(Enum):
    float32 =  "float32"
    Other = "Other"

    def from_str(precision):
        if precision in ["torch.float32", "float32"]:
            return Precision.float32
        return Precision.Other

# Column selection
COLS = [c.name for c in fields(AutoEvalColumn)]

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

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