src_maia/display/utils.py

from dataclasses import dataclass, make_dataclass
from enum import Enum

import pandas as pd

from src_maia.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(["rank", ColumnContent, ColumnContent("Rank", "number", True, never_hidden=True)])
auto_eval_column_dict.append(["size_symbol", ColumnContent, ColumnContent("Size", "number", True, never_hidden=True)])

auto_eval_column_dict.append(["fewshot_symbol", ColumnContent, ColumnContent("FS", "str", True, never_hidden=True)])
auto_eval_column_dict.append(["is_5fewshot", ColumnContent, ColumnContent("IS_FS", "bool", True)])

auto_eval_column_dict.append(["mode_symbol", ColumnContent, ColumnContent("Mode", "str", True, never_hidden=True)])
auto_eval_column_dict.append(["is_multimodal", ColumnContent, ColumnContent("IS_MULTIMODAL", "bool", True)])

auto_eval_column_dict.append(["model", ColumnContent, ColumnContent("Model", "markdown", True, never_hidden=True)])
#auto_eval_column_dict.append(["fewshot", ColumnContent, ColumnContent("Few-Shot", "str", True)])

#Scores
auto_eval_column_dict.append(["average", ColumnContent, ColumnContent("Avg. Comb. Perf. ⬆️", "number", True)])
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", "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)])
#auto_eval_column_dict.append(["submitted_time", ColumnContent, ColumnContent("Submitted time", "date", False)])

# 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", "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="5-few-shot", symbol="5️⃣")
    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 ModeDetails:
    name: str
    symbol: str = ""  # emoji

class ModeType(Enum):
    TEXT = ModeDetails(name="TextOnly", symbol="🔤")
    MULTIMODAL = ModeDetails(name="Multimodal", symbol="🔤🖼️")
    Unknown = ModeDetails(name="unknown", symbol="❓")

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

    @staticmethod
    def from_num_mode(is_multimodal):
        """Determines FewShotType based on num_fewshot."""
        if is_multimodal is False:
            return ModeType.TEXT
        elif is_multimodal is True:
            return ModeType.MULTIMODAL
        return ModeType.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]
'''