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("T", "str", True, never_hidden=True)))
auto_eval_column_dict.append(("model", ColumnContent("Model", "markdown", True, never_hidden=True)))
# Average score
auto_eval_column_dict.append(("average", ColumnContent("Average", "number", True)))
#Scores
for task in Tasks:
    auto_eval_column_dict.append((task.name, ColumnContent(task.value.col_name, "number", True)))
# Model information
auto_eval_column_dict.append(("precision", ColumnContent("Precision", "str", False)))
auto_eval_column_dict.append(("license", ColumnContent("Hub License", "str", False)))
auto_eval_column_dict.append(("params", ColumnContent("#Params (B)", "number", False)))
auto_eval_column_dict.append(("likes", ColumnContent("Hub ❤️", "number", False)))
auto_eval_column_dict.append(("still_on_hub", ColumnContent("Available on the hub", "bool", False)))
auto_eval_column_dict.append(("revision", 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)

## 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)
    precision = ColumnContent("precision", "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):
    FT = ModelDetails(name="fine-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
        return ModelType.Unknown

    @staticmethod
    def from_config(config):
        """Determine model type from configuration - for NER models, most will be fine-tuned"""
        if hasattr(config, 'num_labels') and config.num_labels > 2:
            return ModelType.FT  # Fine-tuned for NER
        return ModelType.Unknown

class WeightType(Enum):
    Original = ModelDetails("Original")

class Precision(Enum):
    float16 = ModelDetails("float16")
    bfloat16 = ModelDetails("bfloat16")
    Unknown = ModelDetails("?")

    @staticmethod
    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]