src/display/utils.py

from dataclasses import dataclass, make_dataclass
from enum import Enum

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
    never_displayed: bool = False

## Leaderboard columns
auto_eval_column_dict = []
# Init
auto_eval_column_dict.append(["model_type_symbol", ColumnContent, ColumnContent("Type Symbol", "str", True, never_hidden=True)])
# auto_eval_column_dict.append(["model", ColumnContent, ColumnContent("Model", "markdown", True, never_hidden=True)])
auto_eval_column_dict.append(["retrieval_model", ColumnContent, ColumnContent("Retrieval Model Plain", "markdown", False, never_displayed=True)])
auto_eval_column_dict.append(["generative_model", ColumnContent, ColumnContent("Generative Model Plain", "markdown", False, never_displayed=True)])
auto_eval_column_dict.append(["retrieval_model_link", ColumnContent, ColumnContent("Retrieval Model", "markdown", True, never_hidden=True)])
auto_eval_column_dict.append(["generative_model_link", ColumnContent, ColumnContent("Generative Model", "markdown", True, never_hidden=True)])

#Scores
auto_eval_column_dict.append(["gen_average", ColumnContent, ColumnContent("Gen Average ⬆️", "number", True)])
auto_eval_column_dict.append(["ret_average", ColumnContent, ColumnContent("Ret Average ⬆️", "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(["gen_num_params", ColumnContent, ColumnContent("Gen#Params (B)", "number", False)])
auto_eval_column_dict.append(["ret_num_params", ColumnContent, ColumnContent("Ret#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)])

# We use make dataclass to dynamically fill the scores from Tasks
AutoEvalColumn = make_dataclass("AutoEvalColumn", auto_eval_column_dict, frozen=True)
EvalQueueColumn = make_dataclass("EvalQueueColumn", 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):
    OpenSource = ModelDetails(name="open-source", symbol="🟢")
    # FT = ModelDetails(name="fine-tuned", symbol="🔶")
    ClosedSource = ModelDetails(name="closed-source", 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 "open-source" in type or "🟢" in type:
            return ModelType.OpenSource
        if "closed-source" in type or "⭕" in type:
            return ModelType.ClosedSource
        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]