LiveBench / src /display /utils.py
pufanyi's picture
Refactor init_leaderboard function to update data outputs to dataframe and improve dropdown UI
f5f1257
raw
history blame
2.18 kB
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, ColumnContent("T", "str", True, never_hidden=True)])
auto_eval_column_dict.append(
["model", ColumnContent, ColumnContent("Model Name", "str", True, True, never_hidden=True)]
)
# Scores
auto_eval_column_dict.append(["Overall", ColumnContent, ColumnContent("Total", "number", True)])
for task in Tasks:
auto_eval_column_dict.append([task.name, ColumnContent, ColumnContent(task.value.col_name, "number", True)])
auto_eval_column_dict.append(["dataset_version", ColumnContent, ColumnContent("Dataset Version", "str", False, True)])
# 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
# 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]