src/display/utils.py

from dataclasses import dataclass, field, make_dataclass
from typing import ClassVar
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_fields = []
# # Init
# auto_eval_column_fields.append(["model_type_symbol", ColumnContent, ColumnContent("T", "str", True, never_hidden=True)])
# auto_eval_column_fields.append(["model", ColumnContent, ColumnContent("Model", "markdown", True, never_hidden=True)])
# # Scores
# auto_eval_column_fields.append(["average", ColumnContent, ColumnContent("Average ⬆️", "number", True)])
# for task in Tasks:
#     auto_eval_column_fields.append([task.name, ColumnContent, ColumnContent(task.value.col_name, "number", True)])
# # Model information
# auto_eval_column_fields.append(["model_type", ColumnContent, ColumnContent("Type", "str", False)])
# auto_eval_column_fields.append(["architecture", ColumnContent, ColumnContent("Architecture", "str", False)])
# auto_eval_column_fields.append(["weight_type", ColumnContent, ColumnContent("Weight type", "str", False, True)])
# auto_eval_column_fields.append(["precision", ColumnContent, ColumnContent("Precision", "str", False)])
# auto_eval_column_fields.append(["license", ColumnContent, ColumnContent("Hub License", "str", False)])
# auto_eval_column_fields.append(["params", ColumnContent, ColumnContent("#Params (B)", "number", False)])
# auto_eval_column_fields.append(["likes", ColumnContent, ColumnContent("Hub ❤️", "number", False)])
# auto_eval_column_fields.append(["still_on_hub", ColumnContent, ColumnContent("Available on the hub", "bool", False)])
# auto_eval_column_fields.append(["revision", ColumnContent, ColumnContent("Model sha", "str", False, False)])
#
#
#
# def make_classvar_dataclass(name: str, spec: list):
#     ns = {"__annotations__": {}}
#     for field_name, field_type, default in spec:
#         # Mark as ClassVar so dataclass doesn't treat it as an instance field
#         ns["__annotations__"][field_name] = ClassVar[field_type]
#         ns[field_name] = default
#     # No instance fields; just class-level descriptors
#     return make_dataclass(name, [], frozen=True, namespace=ns)
#
# # We use make dataclass to dynamically fill the scores from Tasks
# AutoEvalColumn = make_classvar_dataclass("AutoEvalColumn", auto_eval_column_fields)


@dataclass(frozen=True)
class AutoEvalColumn:
    system = ColumnContent("System Name", "markdown", True, never_hidden=True)
    system_type = ColumnContent("System Type", "str", True)
    organization = ColumnContent("Organization", "str", True, never_hidden=True)
    success_rate = ColumnContent("Success Rate (%)", "number", True)
    problems_solved = ColumnContent("Problems Solved", "number", True)
    submitted_on = ColumnContent("Submitted On", "datetime", 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):
    LLM = ModelDetails(name="LLM", symbol="🟢")
    AgenticLLM = ModelDetails(name="AgenticLLM", symbol="🔶")
    # IFT = ModelDetails(name="instruction-tuned", symbol="⭕")
    # RL = ModelDetails(name="RL-tuned", symbol="🟦")
    Other = ModelDetails(name="Other", symbol="?")

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

    @staticmethod
    def from_str(type):
        if "AgenticLLM" in type or "🔶" in type:
            return ModelType.AgenticLLM
        if "LLM" in type or "🟢" in type:
            return ModelType.LLM
        # if "RL-tuned" in type or "🟦" in type:
        #     return ModelType.RL
        # if "instruction-tuned" in type or "⭕" in type:
        #     return ModelType.IFT
        return ModelType.Other


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]