src/display/utils.py

#!/usr/bin/env python
# -*- coding: utf-8 -*-
# flake8: noqa E501

from dataclasses import dataclass, field, make_dataclass
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 not k.startswith("__") and not k.endswith("__")]

# 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 = field(default='str')
    displayed_by_default: bool = field(default=False)
    hidden: bool = field(default=False)
    never_hidden: bool = field(default=False)


# Helper function to create a ColumnContent with default_factory
def create_column_content(name, type='str', displayed_by_default=False, hidden=False, never_hidden=False):
    return field(default_factory=lambda: ColumnContent(name, type, displayed_by_default, hidden, never_hidden))

auto_eval_column_dict = [
    ("model_type_symbol", ColumnContent, create_column_content("", "str", True, never_hidden=True)),
    ("model", ColumnContent, create_column_content("Model", "markdown", True, never_hidden=True)),
    ("average", ColumnContent, create_column_content("Average", "number", True)),
]

# Add task-specific columns
for task in Tasks:
    auto_eval_column_dict.append((task.name, ColumnContent, create_column_content(task.value.col_name, "number", True)))

# Add model information columns
model_info_columns = [
    ("model_type", "Type", "str", False),
    ("architecture", "Architecture", "str", False),
    ("weight_type", "Weight type", "str", False, True),
    ("precision", "Precision", "str", False),
    ("license", "License", "str", False),
    ("params", "Parameters ⚙️", "number", False),
    ("likes", "Likes ❤️", "number", False),
    ("still_on_hub", "Available on the hub", "bool", False),
    ("revision", "Model sha", "str", False, False),
]

for col_name, display_name, col_type, displayed_by_default, *args in model_info_columns:
    hidden = args[0] if args else False
    auto_eval_column_dict.append((col_name, ColumnContent, create_column_content(display_name, col_type, displayed_by_default, hidden)))

# Create the AutoEvalColumn 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="🟦")
    CS = ModelDetails(name="closed-source", symbol="🔒")
    BrainDAO = ModelDetails(name="braindao", 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


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]