src/display/utils.py

from dataclasses import dataclass, make_dataclass, field
from enum import Enum
from functools import partial

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 = []

auto_eval_column_dict.append((
    "rank",
    ColumnContent,
    field(default_factory=lambda: ColumnContent("Rank", "number", True, never_hidden=True))
))

auto_eval_column_dict.append((
    "size_symbol",
    ColumnContent,
    field(default_factory=lambda: ColumnContent("Size", "number", True, never_hidden=True))
))

auto_eval_column_dict.append((
    "fewshot_symbol",
    ColumnContent,
    field(default_factory=lambda: ColumnContent("FS", "str", True, never_hidden=True))
))

auto_eval_column_dict.append((
    "is_5fewshot",
    ColumnContent,
    field(default_factory=lambda: ColumnContent("IS_FS", "bool", True))
))

auto_eval_column_dict.append((
    "LANG",
    ColumnContent,
    field(default_factory=lambda: ColumnContent("LANG", "str", True, never_hidden=True))
))

auto_eval_column_dict.append((
    "model",
    ColumnContent,
    field(default_factory=lambda: ColumnContent("Model", "markdown", True, never_hidden=True))
))

# Scores
auto_eval_column_dict.append((
    "average",
    ColumnContent,
    field(default_factory=lambda: ColumnContent("Avg. Comb. Perf. ⬆️", "number", True))
))

for task in Tasks:
    auto_eval_column_dict.append((
        task.name,
        ColumnContent,
        field(default_factory=lambda t=task: ColumnContent(t.value.col_name, "number", True))
    ))

# Model info
auto_eval_column_dict.append((
    "architecture",
    ColumnContent,
    field(default_factory=lambda: ColumnContent("Architecture", "str", False))
))

auto_eval_column_dict.append((
    "weight_type",
    ColumnContent,
    field(default_factory=lambda: ColumnContent("Weight type", "str", False, True))
))

auto_eval_column_dict.append((
    "license",
    ColumnContent,
    field(default_factory=lambda: ColumnContent("Hub License", "str", False))
))

auto_eval_column_dict.append((
    "params",
    ColumnContent,
    field(default_factory=lambda: ColumnContent("#Params (B)", "number", False))
))

auto_eval_column_dict.append((
    "likes",
    ColumnContent,
    field(default_factory=lambda: ColumnContent("Hub ❤️", "number", False))
))

auto_eval_column_dict.append((
    "still_on_hub",
    ColumnContent,
    field(default_factory=lambda: ColumnContent("Available on the hub", "bool", False))
))

auto_eval_column_dict.append((
    "revision",
    ColumnContent,
    field(default_factory=lambda: ColumnContent("Model sha", "str", False, False))
))

# Create dataclass
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):
    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

@dataclass
class FewShotDetails:
    name: str
    symbol: str = ""  # emoji

class FewShotType(Enum):
    ZS = FewShotDetails(name="zero-shot", symbol="🅾️")
    FS = FewShotDetails(name="10-few-shot", symbol="🔟")
    Unknown = FewShotDetails(name="unknown", symbol="❓")

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

    @staticmethod
    def from_num_fewshot(is_5fewshot):
        """Determines FewShotType based on num_fewshot."""
        if is_5fewshot is False:
            return FewShotType.ZS
        elif is_5fewshot is True:
            return FewShotType.FS
        return FewShotType.Unknown

@dataclass
class SizeDetails:
    name: str
    symbol: str = ""  # emoji

class SizeType(Enum):
    SMALL = SizeDetails(name="small", symbol="🔵")
    MEDIUM = SizeDetails(name="medium", symbol="🔵🔵")
    LARGE = SizeDetails(name="large", symbol="🔵🔵🔵")
    Unknown = SizeDetails(name="unknown", symbol="❓")

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

    @staticmethod
    def num2type(size):
        """Determines FewShotType based on num_fewshot."""
        if size <= 10:
            return SizeType.SMALL
        elif size > 10 and size <= 50:
            return SizeType.MEDIUM
        else:
            return SizeType.LARGE

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]

'''
# Nuovi valori per CPS, AVERAGE, BEST, e ID nella tabella
@dataclass
class NewColumnContent:
    name: str
    type: str
    displayed_by_default: bool
    hidden: bool = False
    never_hidden: bool = False
'''

'''
new_column_dict = []
# Aggiungi CPS, VERAGE, BEST, ID
new_column_dict.append(["CPS", NewColumnContent, NewColumnContent("CPS", "number", True)])
new_column_dict.append(["AVERAGE", NewColumnContent, NewColumnContent("Average ⬆️", "number", True)])
new_column_dict.append(["BEST", NewColumnContent, NewColumnContent("Best Performance", "number", True)])
new_column_dict.append(["ID", NewColumnContent, NewColumnContent("ID", "str", True)])
NewColumn = make_dataclass("NewColumn", new_column_dict, frozen=True)
NEW_COLS = [c.name for c in fields(NewColumn) if not c.hidden]
'''