CPunisher/JavaBench

1from dataclasses import dataclass, make_dataclass2from enum import Enum3 4import pandas as pd5 6from src.about import Tasks7 8def fields(raw_class):9    return [v for k, v in raw_class.__dict__.items() if k[:2] != "__" and k[-2:] != "__"]10 11 12# These classes are for user facing column names,13# to avoid having to change them all around the code14# when a modif is needed15@dataclass16class ColumnContent:17    name: str18    type: str19    displayed_by_default: bool20    hidden: bool = False21    never_hidden: bool = False22 23## Leaderboard columns24auto_eval_column_dict = []25# Init26# auto_eval_column_dict.append(["model_type_symbol", ColumnContent, ColumnContent("T", "str", True, never_hidden=True)])27# auto_eval_column_dict.append(["model", ColumnContent, ColumnContent("Model", "markdown", True, never_hidden=True)])28#Scores29# auto_eval_column_dict.append(["average", ColumnContent, ColumnContent("Average ⬆️", "number", True)])30# auto_eval_column_dict.append(["model_type", ColumnContent, ColumnContent("Type", "str", False)])31# auto_eval_column_dict.append(["architecture", ColumnContent, ColumnContent("Architecture", "str", False)])32# auto_eval_column_dict.append(["weight_type", ColumnContent, ColumnContent("Weight type", "str", False, True)])33# auto_eval_column_dict.append(["precision", ColumnContent, ColumnContent("Precision", "str", False)])34# auto_eval_column_dict.append(["license", ColumnContent, ColumnContent("Hub License", "str", False)])35# auto_eval_column_dict.append(["params", ColumnContent, ColumnContent("#Params (B)", "number", False)])36# auto_eval_column_dict.append(["likes", ColumnContent, ColumnContent("Hub ❤️", "number", False)])37# auto_eval_column_dict.append(["still_on_hub", ColumnContent, ColumnContent("Available on the hub", "bool", False)])38# auto_eval_column_dict.append(["revision", ColumnContent, ColumnContent("Model sha", "str", False, False)])39 40auto_eval_column_dict.append(["model", ColumnContent, ColumnContent("Model", "markdown", True, never_hidden=True)])41auto_eval_column_dict.append(["context", ColumnContent, ColumnContent("Context", "str", True, never_hidden=True)])42auto_eval_column_dict.append(["method", ColumnContent, ColumnContent("Method", "str", True, never_hidden=True)])43auto_eval_column_dict.append(["completion", ColumnContent, ColumnContent("Completion", "number", True, never_hidden=True)])44auto_eval_column_dict.append(["compilation_class_wise", ColumnContent, ColumnContent("Compilation(class)", "number", True, never_hidden=True)])45auto_eval_column_dict.append(["compilation_test_wise", ColumnContent, ColumnContent("Compilation(test)", "number", True, never_hidden=True)])46auto_eval_column_dict.append(["pass_class_wise", ColumnContent, ColumnContent("Pass(class)", "number", True, never_hidden=True)])47auto_eval_column_dict.append(["pass_test_wise", ColumnContent, ColumnContent("Pass(test)", "number", True, never_hidden=True)])48 49# We use make dataclass to dynamically fill the scores from Tasks50AutoEvalColumn = make_dataclass("AutoEvalColumn", auto_eval_column_dict, frozen=True)51 52## For the queue columns in the submission tab53@dataclass(frozen=True)54class EvalQueueColumn:  # Queue column55    model = ColumnContent("model", "markdown", True)56    revision = ColumnContent("revision", "str", True)57    private = ColumnContent("private", "bool", True)58    precision = ColumnContent("precision", "str", True)59    weight_type = ColumnContent("weight_type", "str", "Original")60    status = ColumnContent("status", "str", True)61 62## All the model information that we might need63@dataclass64class ModelDetails:65    name: str66    display_name: str = ""67    symbol: str = "" # emoji68 69 70class ModelType(Enum):71    PT = ModelDetails(name="pretrained", symbol="🟢")72    FT = ModelDetails(name="fine-tuned", symbol="🔶")73    IFT = ModelDetails(name="instruction-tuned", symbol="⭕")74    RL = ModelDetails(name="RL-tuned", symbol="🟦")75    Unknown = ModelDetails(name="", symbol="?")76 77    def to_str(self, separator=" "):78        return f"{self.value.symbol}{separator}{self.value.name}"79 80    @staticmethod81    def from_str(type):82        if "fine-tuned" in type or "🔶" in type:83            return ModelType.FT84        if "pretrained" in type or "🟢" in type:85            return ModelType.PT86        if "RL-tuned" in type or "🟦" in type:87            return ModelType.RL88        if "instruction-tuned" in type or "⭕" in type:89            return ModelType.IFT90        return ModelType.Unknown91 92class WeightType(Enum):93    Adapter = ModelDetails("Adapter")94    Original = ModelDetails("Original")95    Delta = ModelDetails("Delta")96 97class Precision(Enum):98    float16 = ModelDetails("float16")99    bfloat16 = ModelDetails("bfloat16")100    Unknown = ModelDetails("?")101 102    def from_str(precision):103        if precision in ["torch.float16", "float16"]:104            return Precision.float16105        if precision in ["torch.bfloat16", "bfloat16"]:106            return Precision.bfloat16107        return Precision.Unknown108 109# Column selection110COLS = [c.name for c in fields(AutoEvalColumn) if not c.hidden]111 112EVAL_COLS = [c.name for c in fields(EvalQueueColumn)]113EVAL_TYPES = [c.type for c in fields(EvalQueueColumn)]114 115BENCHMARK_COLS = [t.value.col_name for t in Tasks]116 117