FuryAssassin/DebuggedModel-Verified

1# Python fallback for benchmark_utils extracted from compiled module2# Implementations mirror the formulas found in the compiled Cython extension.3 4def calculate_math_reasoning_score(step_value):5    if not isinstance(step_value, int) or step_value <= 0:6        return None7    x = step_value / 100.08    score = 0.3 + 0.5 * (1 - 1/(1 + 0.1*x))9    return round(min(score, 0.95), 3)10 11def calculate_code_generation_score(step_value):12    if not isinstance(step_value, int) or step_value <= 0:13        return None14    x = step_value / 200.015    score = 0.35 + 0.45 * (1 - 1/(1 + 0.08*x))16    return round(min(score, 0.92), 3)17 18def calculate_text_classification_score(step_value):19    if not isinstance(step_value, int) or step_value <= 0:20        return None21    x = step_value / 150.022    score = 0.4 + 0.4 * (1 - 1/(1 + 0.05*x))23    return round(min(score, 0.92), 3)24 25def calculate_sentiment_analysis_score(step_value):26    if not isinstance(step_value, int) or step_value <= 0:27        return None28    x = step_value / 120.029    score = 0.38 + 0.42 * (1 - 1/(1 + 0.04*x))30    return round(min(score, 0.92), 3)31 32def calculate_question_answering_score(step_value):33    if not isinstance(step_value, int) or step_value <= 0:34        return None35    x = step_value / 130.036    score = 0.33 + 0.48 * (1 - 1/(1 + 0.06*x))37    return round(min(score, 0.95), 3)38 39def calculate_logical_reasoning_score(step_value):40    if not isinstance(step_value, int) or step_value <= 0:41        return None42    x = step_value / 110.043    score = 0.42 + 0.4 * (1 - 1/(1 + 0.07*x))44    return round(min(score, 0.95), 3)45 46def calculate_common_sense_score(step_value):47    if not isinstance(step_value, int) or step_value <= 0:48        return None49    x = step_value / 140.050    score = 0.34 + 0.38 * (1 - 1/(1 + 0.05*x))51    return round(min(score, 0.9), 3)52 53def calculate_reading_comprehension_score(step_value):54    if not isinstance(step_value, int) or step_value <= 0:55        return None56    x = step_value / 160.057    score = 0.36 + 0.39 * (1 - 1/(1 + 0.045*x))58    return round(min(score, 0.9), 3)59 60def calculate_dialogue_generation_score(step_value):61    if not isinstance(step_value, int) or step_value <= 0:62        return None63    x = step_value / 170.064    score = 0.31 + 0.45 * (1 - 1/(1 + 0.05*x))65    return round(min(score, 0.9), 3)66 67def calculate_summarization_score(step_value):68    if not isinstance(step_value, int) or step_value <= 0:69        return None70    x = step_value / 180.071    score = 0.45 + 0.35 * (1 - 1/(1 + 0.03*x))72    return round(min(score, 0.9), 3)73 74def calculate_translation_score(step_value):75    if not isinstance(step_value, int) or step_value <= 0:76        return None77    x = step_value / 190.078    score = 0.5 + 0.3 * (1 - 1/(1 + 0.02*x))79    return round(min(score, 0.9), 3)80 81def calculate_knowledge_retrieval_score(step_value):82    if not isinstance(step_value, int) or step_value <= 0:83        return None84    x = step_value / 125.085    score = 0.3 + 0.35 * (1 - 1/(1 + 0.04*x))86    return round(min(score, 0.9), 3)87 88def calculate_creative_writing_score(step_value):89    if not isinstance(step_value, int) or step_value <= 0:90        return None91    x = step_value / 115.092    score = 0.28 + 0.45 * (1 - 1/(1 + 0.06*x))93    return round(min(score, 0.9), 3)94 95def calculate_instruction_following_score(step_value):96    if not isinstance(step_value, int) or step_value <= 0:97        return None98    x = step_value / 135.099    score = 0.37 + 0.44 * (1 - 1/(1 + 0.05*x))100    return round(min(score, 0.95), 3)101 102def calculate_safety_evaluation_score(step_value):103    if not isinstance(step_value, int) or step_value <= 0:104        return None105    x = step_value / 145.0106    score = 0.32 + 0.39 * (1 - 1/(1 + 0.04*x))107    return round(min(score, 0.95), 3)108 109# Mapping110BENCHMARK_CALCULATORS = {111    "math_reasoning": calculate_math_reasoning_score,112    "logical_reasoning": calculate_logical_reasoning_score,113    "code_generation": calculate_code_generation_score,114    "question_answering": calculate_question_answering_score,115    "reading_comprehension": calculate_reading_comprehension_score,116    "common_sense": calculate_common_sense_score,117    "text_classification": calculate_text_classification_score,118    "sentiment_analysis": calculate_sentiment_analysis_score,119    "dialogue_generation": calculate_dialogue_generation_score,120    "summarization": calculate_summarization_score,121    "translation": calculate_translation_score,122    "knowledge_retrieval": calculate_knowledge_retrieval_score,123    "creative_writing": calculate_creative_writing_score,124    "instruction_following": calculate_instruction_following_score,125    "safety_evaluation": calculate_safety_evaluation_score,126}127 128 129def get_benchmark_score(benchmark_name, step_value):130    """Get the score for a specific benchmark given the training step.131    Returns a float or None if invalid.132    """133    calculator = BENCHMARK_CALCULATORS.get(benchmark_name)134    if calculator is None:135        return None136    return calculator(step_value)137