src/constants.py

# reference for length bias categories
length_categories = {
    'alpacaeval-easy': 'True',
    'alpacaeval-hard': 'True',
    'alpacaeval-length': 'Neutral',
    'donotanswer': 'False',
    'hep-cpp': 'Neutral',
    'hep-go': 'Neutral',
    'hep-java': 'Neutral',
    'hep-js': 'Neutral',
    'hep-python': 'Neutral',
    'hep-rust': 'Neutral',
    'llmbar-adver-GPTInst': 'False',
    'llmbar-adver-GPTOut': 'Neutral',
    'llmbar-adver-manual': 'False',
    'llmbar-adver-neighbor': 'False',
    'llmbar-natural': 'Neutral',
    'math-prm': 'Neutral',
    'mt-bench-easy': 'False',
    'mt-bench-hard': 'False',
    'mt-bench-med': 'Neutral',
    'refusals-dangerous': 'False',
    'refusals-offensive': 'False',
    'xstest-should-refuse': 'False',
    'xstest-should-respond': 'True'
}

example_counts = {
    "alpacaeval-easy": 100,
    "alpacaeval-length": 95,
    "alpacaeval-hard": 95,
    "mt-bench-easy": 28,
    "mt-bench-med": 40,
    "mt-bench-hard": 37,
    "math-prm": 984, # actual length 447, upweighting to be equal to code
    "refusals-dangerous": 100,
    "refusals-offensive": 100,
    "llmbar-natural": 100,
    "llmbar-adver-neighbor": 134,
    "llmbar-adver-GPTInst": 92,
    "llmbar-adver-GPTOut": 47,
    "llmbar-adver-manual": 46,
    "xstest-should-refuse": 250,
    "xstest-should-respond": 154,
    "donotanswer": 136,
    "hep-cpp": 164,
    "hep-go": 164,
    "hep-java": 164,
    "hep-js": 164,
    "hep-python": 164,
    "hep-rust": 164
}

# note, this order should match the dataframe.
subset_mapping = {
    "Chat": ['alpacaeval-easy', 'alpacaeval-hard', 'alpacaeval-length', 'mt-bench-easy', 'mt-bench-med'],
    "Chat Hard": ['llmbar-adver-GPTInst', 'llmbar-adver-GPTOut', 'llmbar-adver-manual', 'llmbar-adver-neighbor', 'llmbar-natural', 'mt-bench-hard'],
    "Safety": ['donotanswer', 'refusals-dangerous', 'refusals-offensive', 'xstest-should-refuse', 'xstest-should-respond'],
    "Reasoning": ["hep-cpp", "hep-go", "hep-java", "hep-js", "hep-python", "hep-rust", "math-prm"]
}