Datasets:

mbpp

like 95

23 instructions in the sanitized train set (but 0 in the test set!) contain links, for example:

Some of these include links that contain key information about the task, such as:

{'source_file': 'Benchmark Questions Verification V2.ipynb',
 'task_id': 765,
 'prompt': 'Write a function to find nth polite number. geeksforgeeks.org/n-th-polite-number/',
 'code': 'import math \ndef is_polite(n): \n\tn = n + 1\n\treturn (int)(n+(math.log((n + math.log(n, 2)), 2))) ',
 'test_imports': array([], dtype=object),
 'test_list': array(['assert is_polite(7) == 11', 'assert is_polite(4) == 7',
        'assert is_polite(9) == 13'], dtype=object)}

Some others have links which are just redundant:

{'source_file': 'Benchmark Questions Verification V2.ipynb',
 'task_id': 745,
 'prompt': 'Write a function to find numbers within a given range from startnum ti endnum where every number is divisible by every digit it contains. https://www.w3resource.com/python-exercises/lambda/python-lambda-exercise-24.php',
 'code': 'def divisible_by_digits(startnum, endnum):\n    return [n for n in range(startnum, endnum+1) \\\n                if not any(map(lambda x: int(x) == 0 or n%int(x) != 0, str(n)))]',
 'test_imports': array([], dtype=object),
 'test_list': array(['assert divisible_by_digits(1,22)==[1, 2, 3, 4, 5, 6, 7, 8, 9, 11, 12, 15, 22]',
        'assert divisible_by_digits(1,15)==[1, 2, 3, 4, 5, 6, 7, 8, 9, 11, 12, 15]',
        'assert divisible_by_digits(20,25)==[22, 24]'], dtype=object)}

It may be worth noting in the dataset card where such links come from, or even make it easy to filter them out.