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", "str", True, never_hidden=True)])
auto_eval_column_dict.append(["model", ColumnContent, ColumnContent("Model", "markdown", True, never_hidden=True)])
#Scores
auto_eval_column_dict.append(["average", ColumnContent, ColumnContent("Média Geral ⬆️", "number", True)])

# Média específica do grupo PLUE
auto_eval_column_dict.append(["plue_avg", ColumnContent, ColumnContent("PLUE", "number", True)])

# Adicionando colunas para as médias das áreas (manter para cálculo e outras abas)
auto_eval_column_dict.append(["area_medica_avg", ColumnContent, ColumnContent("Área Médica", "number", False)]) # Exibido por padrão = False
auto_eval_column_dict.append(["area_direito_avg", ColumnContent, ColumnContent("Área do Direito", "number", False)]) # Exibido por padrão = False
auto_eval_column_dict.append(["provas_militares_avg", ColumnContent, ColumnContent("Provas Militares", "number", False)]) # Exibido por padrão = False
auto_eval_column_dict.append(["computacao_avg", ColumnContent, ColumnContent("Computação", "number", False)]) # Exibido por padrão = False
auto_eval_column_dict.append(["discurso_odio_avg", ColumnContent, ColumnContent("Discurso de Ódio", "number", False)]) # Mover para PLUE -> False
auto_eval_column_dict.append(["economia_contabilidade_avg", ColumnContent, ColumnContent("Economia e Contabilidade", "number", False)]) # Mover para PLUE -> False
auto_eval_column_dict.append(["semantica_inferencia_avg", ColumnContent, ColumnContent("Semântica e Inferência", "number", False)]) # Mover para PLUE -> False
auto_eval_column_dict.append(["multidisciplinar_avg", ColumnContent, ColumnContent("Multidisciplinar", "number", False)]) # Exibido por padrão = False
# Médias Novas Áreas
auto_eval_column_dict.append(["energy_avg", ColumnContent, ColumnContent("Energy", "number", False)]) # PLUE -> False
auto_eval_column_dict.append(["reasoning_avg", ColumnContent, ColumnContent("Reasoning", "number", False)]) # PLUE -> False

for task in Tasks:
    auto_eval_column_dict.append([task.name, ColumnContent, ColumnContent(task.value.col_name, "number", False)])
# Model information
auto_eval_column_dict.append(["model_type", ColumnContent, ColumnContent("Type", "str", False)])
auto_eval_column_dict.append(["architecture", ColumnContent, ColumnContent("Architecture", "str", False)])
auto_eval_column_dict.append(["weight_type", ColumnContent, ColumnContent("Weight type", "str", False, True)])
auto_eval_column_dict.append(["precision", ColumnContent, ColumnContent("Precision", "str", False)])
auto_eval_column_dict.append(["license", ColumnContent, ColumnContent("Hub License", "str", False)])
auto_eval_column_dict.append(["params", ColumnContent, ColumnContent("#Params (B)", "number", False)])
auto_eval_column_dict.append(["likes", ColumnContent, ColumnContent("Hub ❤️", "number", False)])
auto_eval_column_dict.append(["still_on_hub", ColumnContent, ColumnContent("Available on the hub", "bool", False)])
auto_eval_column_dict.append(["revision", ColumnContent, ColumnContent("Model sha", "str", False, False)])

# We use make dataclass to dynamically fill the scores from Tasks
AutoEvalColumn = make_dataclass("AutoEvalColumn", auto_eval_column_dict, frozen=True)

# Mapeamento das áreas de conhecimento para os Tasks correspondentes
AREA_DEFINITIONS = {
    "Área Médica": [Tasks.REVALIDA, Tasks.MREX],
    "Área do Direito": [Tasks.OAB, Tasks.ENAM],
    "Provas Militares": [Tasks.AFA, Tasks.ITA, Tasks.IME],
    "Computação": [Tasks.POSCOMP, Tasks.OBI],
    "Discurso de Ódio": [Tasks.HATEBR, Tasks.PT_HATE_SPEECH, Tasks.TWEETSENTBR],
    "Economia e Contabilidade": [Tasks.BCB, Tasks.CFCES],
    "Semântica e Inferência": [Tasks.FAQUAD_NLI, Tasks.ASSIN2_RTE, Tasks.ASSIN2_STS],
    "Multidisciplinar": [Tasks.ENEM, Tasks.BLUEX, Tasks.CNPU, Tasks.ENADE, Tasks.BNDES, Tasks.CACD_1, Tasks.CACD_2],
    # Novas Áreas
    "Energy": [Tasks.ENERGY_DATASET],
    "Reasoning": [Tasks.REASONING_DATASET],
}

# Mapeamento dos nomes das áreas para as colunas de média correspondentes (Manter todos)
AREA_AVG_COLUMN_MAP = {
    "Área Médica": AutoEvalColumn.area_medica_avg.name,
    "Área do Direito": AutoEvalColumn.area_direito_avg.name,
    "Provas Militares": AutoEvalColumn.provas_militares_avg.name,
    "Computação": AutoEvalColumn.computacao_avg.name,
    "Discurso de Ódio": AutoEvalColumn.discurso_odio_avg.name,
    "Economia e Contabilidade": AutoEvalColumn.economia_contabilidade_avg.name,
    "Semântica e Inferência": AutoEvalColumn.semantica_inferencia_avg.name,
    "Multidisciplinar": AutoEvalColumn.multidisciplinar_avg.name,
    # Novas Áreas
    "Energy": AutoEvalColumn.energy_avg.name,
    "Reasoning": AutoEvalColumn.reasoning_avg.name,
}

# --- Definição do Grupo PLUE Atualizado ---
PLUE_GROUP_AREAS = [
    "Área Médica",
    "Área do Direito",
    "Provas Militares",
    "Computação",
    "Discurso de Ódio",
    "Economia e Contabilidade",
    "Semântica e Inferência",
    "Multidisciplinar"
]
# -------

## 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):
    PT = ModelDetails(name="Pre trained", symbol="🟢")
    SFT = ModelDetails(name="Supervised Finetuning", symbol="🔶")
    RL = ModelDetails(name="Reinforcement Learning", symbol="🟦")
    Unknown = ModelDetails(name="", symbol="?")

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

    @staticmethod
    def from_str(type_str):
        if "fine-tuned" in type_str.lower() or \
           "instruction-tuned" in type_str.lower() or \
           "supervised finetuning" in type_str.lower() or \
           "🔶" in type_str or \
           type_str == "SFT" or type_str == "FT" or type_str == "IFT":
            return ModelType.SFT
        if "pretrained" in type_str.lower() or "pre trained" in type_str.lower() or "pré-treinado" in type_str.lower() or "🟢" in type_str or type_str == "PT":
            return ModelType.PT
        if "rl-tuned" in type_str.lower() or "reinforcement learning" in type_str.lower() or "🟦" in type_str or type_str == "RL":
            return ModelType.RL
        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)]

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