question_id stringlengths 32 32 | category stringclasses 11
values | sub_category stringclasses 110
values | turns listlengths 1 5 | source stringclasses 16
values | src_id stringlengths 3 10.6k | difficulty stringclasses 3
values | multiturn bool 2
classes |
|---|---|---|---|---|---|---|---|
0daf539b787c4dccbb547330a8b4c3d7 | coding | null | [
"import (\n \"strings\"\n)\n\n// For a given string, flip lowercase characters to uppercase and uppercase to lowercase.\n// >>> FlipCase('Hello')\n// 'hELLO'\nfunc FlipCase(str string) string {\n"
] | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Go/27 | null | false |
135c7fe91faa48fd83ca5eac94c09f00 | coding | null | [
"\ndef valid_date(date):\n \"\"\"You have to write a function which validates a given date string and\n returns True if the date is valid otherwise False.\n The date is valid if all of the following rules are satisfied:\n 1. The date string is not empty.\n 2. The number of days is not less than 1 or ... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Python/124 | null | false |
37f349607a3d46b8bc5aafcefb5fd518 | coding | null | [
"/*You are given a word. Your task is to find the closest vowel that stands between \n two consonants from the right side of the word (case sensitive).\n \n Vowels in the beginning and ending doesn't count. Return empty string if you didn't\n find any vowel met the above condition. \n\n You may assume that the... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | JavaScript/118 | null | false |
1fd6d82bbde647f29b93afd7b4eb14b0 | coding | null | [
"Write a simple website in HTML. When a user clicks the button, it shows a random joke from a list of 4 jokes.",
"How to use CSS to change the color of jokes to red?"
] | https://raw.githubusercontent.com/hemingkx/Spec-Bench/66230f10cb0a02aced5ef3ce1e85163c16160454/data/spec_bench/question.jsonl | spec_bench_42 | medium | true |
48579d93b98d459fbf473748a607d748 | coding | null | [
"from typing import List\n\n\ndef parse_nested_parens(paren_string: str) -> List[int]:\n \"\"\" Input to this function is a string represented multiple groups for nested parentheses separated by spaces.\n For each of the group, output the deepest level of nesting of parentheses.\n E.g. (()()) has maximum t... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Python/6 | null | false |
5ace1d46a306452da3c517a7a8bc2093 | coding | null | [
"Implement a function to find the median of two sorted arrays of different sizes with O(1) space complexity and O(n) time complexity.",
"Does there exist an implementation with better time complexity?"
] | https://raw.githubusercontent.com/hemingkx/Spec-Bench/66230f10cb0a02aced5ef3ce1e85163c16160454/data/spec_bench/question.jsonl | spec_bench_45 | medium | true |
3416d0ce7c7a4632943ff1326913437f | coding | null | [
"/*\nGiven a string 'text\", return its md5 hash equivalent string.\nIf 'text\" is an empty string, return None.\n\n>>> string_to_md5(\"Hello world\") == \"3e25960a79dbc69b674cd4ec67a72c62\"\n*/\n#include<stdio.h>\n#include<string>\n#include<openssl/md5.h>\nusing namespace std;\nstring string_to_md5(string text){\n... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | CPP/162 | null | false |
e281ab7c608141a683519a317e998dd2 | coding | null | [
"\ndef intersection(interval1, interval2):\n \"\"\"You are given two intervals,\n where each interval is a pair of integers. For example, interval = (start, end) = (1, 2).\n The given intervals are closed which means that the interval (start, end)\n includes both start and end.\n For each given inter... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Python/127 | null | false |
20de4633029f426fbd0c3c4eb6bd3853 | coding | null | [
"\n// Given list of numbers (of at least two elements), apply a linear transform to that list,\n// such that the smallest number will become 0 and the largest will become 1\n// >>> RescaleToUnit([1.0, 2.0, 3.0, 4.0, 5.0])\n// [0.0, 0.25, 0.5, 0.75, 1.0]\nfunc RescaleToUnit(numbers []float64) []float64 {\n"
] | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Go/21 | null | false |
988e25df0fba43039adf10baf818f7a9 | coding | null | [
"\ndef Strongest_Extension(class_name, extensions):\n \"\"\"You will be given the name of a class (a string) and a list of extensions.\n The extensions are to be used to load additional classes to the class. The\n strength of the extension is as follows: Let CAP be the number of the uppercase\n letters ... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Python/153 | null | false |
0fe036855bd4410c99eca693e49248cc | coding | null | [
"A binary tree is full if all of its vertices have either zero or two children. Let B_n denote the number of full binary trees with n vertices. Implement a function to find B_n.",
"What if the problem changed from a binary tree to a ternary tree?"
] | https://raw.githubusercontent.com/hemingkx/Spec-Bench/66230f10cb0a02aced5ef3ce1e85163c16160454/data/spec_bench/question.jsonl | spec_bench_47 | medium | true |
91ec653700af4857857a159f8c6da4c7 | coding | null | [
"\ndef closest_integer(value):\n '''\n Create a function that takes a value (string) representing a number\n and returns the closest integer to it. If the number is equidistant\n from two integers, round it away from zero.\n\n Examples\n >>> closest_integer(\"10\")\n 10\n >>> closest_integer... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Python/99 | null | false |
154d2a4f6556463793035c90e4250561 | coding | null | [
"import java.util.*;\nimport java.lang.*;\n\nclass Solution {\n /**\n You are given a 2 dimensional data, as a nested lists,\n which is similar to matrix, however, unlike matrices,\n each row may contain a different number of columns.\n Given lst, and integer x, find integers x in the list,\n and ... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Java/87 | null | false |
22f2c7dc2f7745e2857160543c2cf401 | coding | null | [
"\ndef check_dict_case(dict):\n \"\"\"\n Given a dictionary, return True if all keys are strings in lower \n case or all keys are strings in upper case, else return False.\n The function should return False is the given dictionary is empty.\n Examples:\n check_dict_case({\"a\":\"apple\", \"b\":\"b... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Python/95 | null | false |
80156ace23984b74891e02065411ef87 | coding | null | [
"fn main(){}\n\nuse std::{slice::Iter, cmp::{max, self}, mem::replace, collections::{HashSet, HashMap}, ops::Index, ascii::AsciiExt};\nuse rand::Rng;\nuse regex::Regex;\nuse md5;\nuse std::any::{Any, TypeId};\n\nfn encode_shift(s: &str) -> String {\n\n let alphabet:Vec<&str> = vec![\"a\", \"b\", \"c\", \"d\", \"... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Rust/50 | null | false |
60868b6276204b9ba3e59ad45067b7fd | coding | null | [
"Develop a Python program that reads all the text files under a directory and returns top-5 words with the most number of occurrences.",
"Can you parallelize it?"
] | https://raw.githubusercontent.com/hemingkx/Spec-Bench/66230f10cb0a02aced5ef3ce1e85163c16160454/data/spec_bench/question.jsonl | spec_bench_40 | medium | true |
cd4f5c325d534a3cb5e243c016f8f3b9 | coding | null | [
"/*You have to write a function which validates a given date string and\n returns true if the date is valid otherwise false.\n The date is valid if all of the following rules are satisfied:\n 1. The date string is not empty.\n 2. The number of days is not less than 1 or higher than 31 days for months 1,3,5,7,8,... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | JavaScript/124 | null | false |
4b328c08db884b959183f68428cc02cc | coding | null | [
"\ndef bf(planet1, planet2):\n '''\n There are eight planets in our solar system: the closerst to the Sun \n is Mercury, the next one is Venus, then Earth, Mars, Jupiter, Saturn, \n Uranus, Neptune.\n Write a function that takes two planet names as strings planet1 and planet2. \n The function shou... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Python/148 | null | false |
21a642e08bdc4040803897dae92c3b6e | coding | null | [
"Write a C++ program to find the nth Fibonacci number using recursion.",
"Now we define a sequence of numbers in which each number is the sum of the three preceding ones. The first three numbers are 0, -1, -1. Write a program to find the nth number."
] | https://raw.githubusercontent.com/hemingkx/Spec-Bench/66230f10cb0a02aced5ef3ce1e85163c16160454/data/spec_bench/question.jsonl | spec_bench_41 | medium | true |
7635d4e93e514bf6bce51b2416b2c1d9 | coding | null | [
"\ndef fix_spaces(text):\n \"\"\"\n Given a string text, replace all spaces in it with underscores, \n and if a string has more than 2 consecutive spaces, \n then replace all consecutive spaces with - \n \n fix_spaces(\"Example\") == \"Example\"\n fix_spaces(\"Example 1\") == \"Example_1\"\n ... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Python/140 | null | false |
b9ec7a3ce76342ed80071dac82c09678 | coding | null | [
"import (\n \"math\"\n)\n\n// You are given an array arr of integers and you need to return\n// sum of magnitudes of integers multiplied by product of all signs\n// of each number in the array, represented by 1, -1 or 0.\n// Note: return nil for empty arr.\n// \n// Example:\n// >>> ProdSigns([1, 2, 2, -4]) == -9... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Go/128 | null | false |
7ea2cb7770904ea3b78f43f466748b6c | coding | null | [
"FULL BENCHMARK DATA SHOULD BE FETCHED FROM THE SOURCE USING SPECDEC_BENCH\n\nYou will be given a problem specification and will generate a correct java program, that matches the specification and passes all tests.\n\n{question}\n\n"
] | https://huggingface.co/datasets/deepmind/code_contests/tree/802411c3010cb00d1b05bad57ca77365a3c699d6 | 1575_B. Building an Amusement Park | easy | false |
8e3677237d6f4ba3b24298f75709dd0b | coding | null | [
"/*\nInput are two strings a and b consisting only of 1s and 0s.\nPerform binary XOR on these inputs and return result also as a string.\n>>> string_xor(\"010\", \"110\")\n\"100\"\n*/\n#include<stdio.h>\n#include<string>\nusing namespace std;\nstring string_xor(string a,string b){\n"
] | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | CPP/11 | null | false |
f5bc6e1054a14617875475b0472fa018 | coding | null | [
"/*Create a function which takes a string representing a file's name, and returns\n 'Yes' if the the file's name is valid, and returns 'No' otherwise.\n A file's name is considered to be valid if and only if all the following conditions \n are met:\n - There should not be more than three digits ('0'-'9') in the... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | JavaScript/141 | null | false |
ecf92ab7acf1492da70e68fcddd9d0b7 | coding | null | [
"\ndef minPath(grid, k):\n \"\"\"\n Given a grid with N rows and N columns (N >= 2) and a positive integer k, \n each cell of the grid contains a value. Every integer in the range [1, N * N]\n inclusive appears exactly once on the cells of the grid.\n\n You have to find the minimum path of length k i... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Python/129 | null | false |
7df494fd138345ff93caadc693a0b1b0 | coding | null | [
"from typing import List\n\n\ndef separate_paren_groups(paren_string: str) -> List[str]:\n \"\"\" Input to this function is a string containing multiple groups of nested parentheses. Your goal is to\n separate those group into separate strings and return the list of those.\n Separate groups are balanced (e... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Python/1 | null | false |
97f49a2c2ad14e7d8cf106551b593c69 | coding | null | [
"Write a function to find the highest common ancestor (not LCA) of two nodes in a binary tree.",
"What if it is not a binary tree?"
] | https://raw.githubusercontent.com/hemingkx/Spec-Bench/66230f10cb0a02aced5ef3ce1e85163c16160454/data/spec_bench/question.jsonl | spec_bench_44 | medium | true |
22f088e74c284523b965617fa401840a | coding | null | [
"import (\n \"sort\"\n)\n\n// Given an array of non-negative integers, return a copy of the given array after sorting,\n// you will sort the given array in ascending order if the sum( first index value, last index value) is odd,\n// or sort it in descending order if the sum( first index value, last index value) ... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Go/88 | null | false |
a6acc9365c464865ad61f02f8b2360aa | coding | null | [
"FULL BENCHMARK DATA SHOULD BE FETCHED FROM THE SOURCE USING SPECDEC_BENCH\n\nYou will be given a problem specification and will generate a correct python program, that matches the specification and passes all tests.\n\n{question}\n\n"
] | https://huggingface.co/datasets/deepmind/code_contests/tree/802411c3010cb00d1b05bad57ca77365a3c699d6 | 1580_E. Railway Construction | easy | false |
c848ad5326ce4238bdc30b6ab0343b92 | coding | null | [
"import math\n\n\ndef poly(xs: list, x: float):\n \"\"\"\n Evaluates polynomial with coefficients xs at point x.\n return xs[0] + xs[1] * x + xs[1] * x^2 + .... xs[n] * x^n\n \"\"\"\n return sum([coeff * math.pow(x, i) for i, coeff in enumerate(xs)])\n\n\ndef find_zero(xs: list):\n \"\"\" xs are c... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Python/32 | null | false |
5126ccb16c2146b5876d6a7da63e9fc1 | coding | null | [
"\n// Check if two words have the same characters.\n// >>> SameChars('eabcdzzzz', 'dddzzzzzzzddeddabc')\n// true\n// >>> SameChars('abcd', 'dddddddabc')\n// true\n// >>> SameChars('dddddddabc', 'abcd')\n// true\n// >>> SameChars('eabcd', 'dddddddabc')\n// false\n// >>> SameChars('abcd', 'dddddddabce')\n// false\n//... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Go/54 | null | false |
709706096071415b9f3bd1d558693d1f | coding | null | [
"\ndef int_to_mini_roman(number):\n \"\"\"\n Given a positive integer, obtain its roman numeral equivalent as a string,\n and return it in lowercase.\n Restrictions: 1 <= num <= 1000\n\n Examples:\n >>> int_to_mini_roman(19) == 'xix'\n >>> int_to_mini_roman(152) == 'clii'\n >>> int_to_mini_r... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Python/156 | null | false |
b76c05c683154ddc867e4d85ec7dde3e | coding | null | [
"/*You will be given the name of a class (a string) and a list of extensions.\n The extensions are to be used to load additional classes to the class. The\n strength of the extension is as follows: Let CAP be the number of the uppercase\n letters in the extension's name, and let SM be the number of lowercase let... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | JavaScript/153 | null | false |
dc8c14e68d7a4de59c48ad5d412e8bbb | coding | null | [
"\n\ndef prime_fib(n: int):\n \"\"\"\n prime_fib returns n-th number that is a Fibonacci number and it's also prime.\n >>> prime_fib(1)\n 2\n >>> prime_fib(2)\n 3\n >>> prime_fib(3)\n 5\n >>> prime_fib(4)\n 13\n >>> prime_fib(5)\n 89\n \"\"\"\n"
] | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Python/39 | null | false |
90f41b6c201149f0846516c2221552a5 | coding | null | [
"Write a function to find the majority element in a given integer array using the Boyer-Moore Voting Algorithm.",
"How about finding the top-2 most occurring elements?"
] | https://raw.githubusercontent.com/hemingkx/Spec-Bench/66230f10cb0a02aced5ef3ce1e85163c16160454/data/spec_bench/question.jsonl | spec_bench_46 | medium | true |
182e7a241cd64f309a013f89f2dde604 | coding | null | [
"\ndef match_parens(lst):\n '''\n You are given a list of two strings, both strings consist of open\n parentheses '(' or close parentheses ')' only.\n Your job is to check if it is possible to concatenate the two strings in\n some order, that the resulting string will be good.\n A string S is cons... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Python/119 | null | false |
a58f4fb5e86543e9862b62598fa4913b | coding | null | [
"FULL BENCHMARK DATA SHOULD BE FETCHED FROM THE SOURCE USING SPECDEC_BENCH\n\nYou will be given a problem specification and will generate a correct cpp program, that matches the specification and passes all tests.\n\n{question}\n\n"
] | https://huggingface.co/datasets/deepmind/code_contests/tree/802411c3010cb00d1b05bad57ca77365a3c699d6 | 1613_C. Poisoned Dagger | easy | false |
de6fabc20e3b4e1db4b039330d57e641 | coding | null | [
"/*\n Write a function that takes a message, and encodes in such a \n way that it swaps case of all letters, replaces all vowels in \n the message with the letter that appears 2 places ahead of that \n vowel in the english alphabet. \n Assume only letters. \n \n Examples:\n >>> encode('test')\n 'TGST'\n >... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | JavaScript/93 | null | false |
76dfa9e332124de5a5403fcf51e61050 | coding | null | [
"import (\n \"math\"\n)\n\n// For a given list of input numbers, calculate Mean Absolute Deviation\n// around the mean of this dataset.\n// Mean Absolute Deviation is the average absolute difference between each\n// element and a centerpoint (mean in this case):\n// MAD = average | x - x_mean |\n// >>> MeanAbsol... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Go/4 | null | false |
f2bf4c564d0d42a39241e98a4e6c493f | coding | null | [
"#include<stdio.h>\n#include<string>\nusing namespace std;\nbool is_palindrome(string str){\n //Test if given string is a palindrome \n string s(str.rbegin(),str.rend());\n return s==str;\n}\nstring make_palindrome(string str){\n /*\n Find the shortest palindrome that begins with a supplied string. \... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | CPP/10 | null | false |
38603327dbd2472a93a652ce7706ef78 | coding | null | [
"\ndef file_name_check(file_name):\n \"\"\"Create a function which takes a string representing a file's name, and returns\n 'Yes' if the the file's name is valid, and returns 'No' otherwise.\n A file's name is considered to be valid if and only if all the following conditions \n are met:\n - There sh... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Python/141 | null | false |
08a5bcca5621439cb1b93a663843d75c | coding | null | [
"FULL BENCHMARK DATA SHOULD BE FETCHED FROM THE SOURCE USING SPECDEC_BENCH\n\nYou will be given a question (problem specification) and will generate a correct program, in java programming language, that matches the specification and passes all tests.\n\nQuestion: {question}\n\nRead the inputs from stdin, solve the ... | https://huggingface.co/datasets/livecodebench/code_generation_lite/tree/0fe84c3912ea0c4d4a78037083943e8f0c4dd505 | abc370_d | medium | false |
42c1487f04e648fdaeb04aed702d05ea | coding | null | [
"import java.util.*;\nimport java.lang.*;\n\nclass Solution {\n /**\n Given an array of integers nums, find the minimum sum of any non-empty sub-array\n of nums.\n Example\n minSubArraySum(Arrays.asList(2, 3, 4, 1, 2, 4)) == 1\n minSubArraySum(Arrays.asList(-1, -2, -3)) == -6\n */\n public ... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Java/114 | null | false |
0e608dd6d8794d0aa40b1a1b2748b699 | coding | null | [
"/* Input to this function is a string containing multiple groups of nested parentheses. Your goal is to\n separate those group into separate strings and return the list of those.\n Separate groups are balanced (each open brace is properly closed) and not nested within each other\n Ignore any spaces in the input... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | JavaScript/1 | null | false |
bba6da07402a44e59a444de6f6b672d4 | coding | null | [
"import (\n \"bytes\"\n)\n\n// You are given a word. Your task is to find the closest vowel that stands between\n// two consonants from the right side of the word (case sensitive).\n// \n// Vowels in the beginning and ending doesn't count. Return empty string if you didn't\n// find any vowel met the above condit... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Go/118 | null | false |
18c06caa12064698a1f7377225b37aa2 | coding | null | [
"\n\ndef largest_prime_factor(n: int):\n \"\"\"Return the largest prime factor of n. Assume n > 1 and is not a prime.\n >>> largest_prime_factor(13195)\n 29\n >>> largest_prime_factor(2048)\n 2\n \"\"\"\n"
] | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Python/59 | null | false |
e967915d991c4cfe96e56cddc8acbc8e | coding | null | [
"/*Task\n We are given two strings s and c, you have to deleted all the characters in s that are equal to any character in c\n then check if the result string is palindrome.\n A string is called palindrome if it reads the same backward as forward.\n You should return a tuple containing the result string and tru... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | JavaScript/112 | null | false |
72556488e2be47fe9980caa01e3fef35 | coding | null | [
"FULL BENCHMARK DATA SHOULD BE FETCHED FROM THE SOURCE USING SPECDEC_BENCH\n\nYou will be given a question (problem specification) and will generate a correct program, in python programming language, that matches the specification and passes all tests.\n\nQuestion: {question}\n\nRead the inputs from stdin, solve th... | https://huggingface.co/datasets/livecodebench/code_generation_lite/tree/0fe84c3912ea0c4d4a78037083943e8f0c4dd505 | abc397_a | easy | false |
eaa1ffd6d1b04beb91923a028c426db3 | coding | null | [
"\n// Given a string text, replace all spaces in it with underscores,\n// and if a string has more than 2 consecutive spaces,\n// then replace all consecutive spaces with -\n// \n// FixSpaces(\"Example\") == \"Example\"\n// FixSpaces(\"Example 1\") == \"Example_1\"\n// FixSpaces(\" Example 2\") == \"_Example_2\"\n/... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Go/140 | null | false |
08f4c31730b4427f90e8146fef066c4c | coding | null | [
"from typing import List, Tuple\n\n\ndef find_closest_elements(numbers: List[float]) -> Tuple[float, float]:\n \"\"\" From a supplied list of numbers (of length at least two) select and return two that are the closest to each\n other and return them in order (smaller number, larger number).\n >>> find_clos... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Python/20 | null | false |
2c3dda5236e14337b8a5cca80754f0a2 | coding | null | [
"import java.util.*;\nimport java.lang.*;\n\nclass Solution {\n /**\n You have to write a function which validates a given date string and\n returns true if the date is valid otherwise false.\n The date is valid if all of the following rules are satisfied:\n 1. The date string is not empty.\n 2. T... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Java/124 | null | false |
11713dbc6908422a8fb942ec607cb850 | coding | null | [
"/*\n Given two lists operator, and operand. The first list has basic algebra operations, and \n the second list is a list of integers. Use the two given lists to build the algebric \n expression and return the evaluation of this expression.\n\n The basic algebra operations:\n Addition ( + ) \n Subtraction ( ... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | JavaScript/160 | null | false |
bca053d2f77e4f918fe652c195402919 | coding | null | [
"\ndef words_in_sentence(sentence):\n \"\"\"\n You are given a string representing a sentence,\n the sentence contains some words separated by a space,\n and you have to return a string that contains the words from the original sentence,\n whose lengths are prime numbers,\n the order of the words ... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Python/143 | null | false |
c486d8856c1544af9e673cc9ee53c8b5 | coding | null | [
"fn main(){}\n\nuse std::{slice::Iter, cmp::{max, self}, mem::replace, collections::{HashSet, HashMap}, ops::Index, ascii::AsciiExt};\nuse rand::Rng;\nuse regex::Regex;\nuse md5;\nuse std::any::{Any, TypeId};\n\npub fn encode_cyclic(s: &str) -> String {\n // returns encoded string by cycling groups of three char... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Rust/38 | null | false |
fa27803acfbf4e55ae341d7624b515f7 | coding | null | [
"FULL BENCHMARK DATA SHOULD BE FETCHED FROM THE SOURCE USING SPECDEC_BENCH\n\nYou will be given a problem specification and will generate a correct java program, that matches the specification and passes all tests.\n\n{question}\n\n"
] | https://huggingface.co/datasets/deepmind/code_contests/tree/802411c3010cb00d1b05bad57ca77365a3c699d6 | 1586_H. Omkar and Tours | easy | false |
1dad57f7698846fdb28cf2ba7631f181 | coding | null | [
"import (\n \"sort\"\n)\n\n// You are given a 2 dimensional data, as a nested lists,\n// which is similar to matrix, however, unlike matrices,\n// each row may contain a different number of columns.\n// Given lst, and integer x, find integers x in the list,\n// and return list of tuples, [(x1, y1), (x2, y2) ...]... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Go/87 | null | false |
c72329fe893f4158b58d9fb6576c19ef | coding | null | [
"/*\n Given a dictionary, return true if all keys are strings in lower \n case or all keys are strings in upper case, else return false.\n The function should return false is the given dictionary is empty.\n Examples:\n checkDictCase({\"a\":\"apple\", \"b\":\"banana\"}) should return true.\n checkDictCase({\"... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | JavaScript/95 | null | false |
c49bd7aa449844da8c67a1bb6797314e | coding | null | [
"\ndef numerical_letter_grade(grades):\n \"\"\"It is the last week of the semester and the teacher has to give the grades\n to students. The teacher has been making her own algorithm for grading.\n The only problem is, she has lost the code she used for grading.\n She has given you a list of GPAs for so... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Python/81 | null | false |
9df42173e2c54dcdb6e241fcc2e3f186 | coding | null | [
"You are given two sorted lists of size m and n. Implement a function to find the kth smallest element in the union of the two lists with linear complexity.",
"Does there exist an algorithm with better time complexity? If so, implement it."
] | https://raw.githubusercontent.com/hemingkx/Spec-Bench/66230f10cb0a02aced5ef3ce1e85163c16160454/data/spec_bench/question.jsonl | spec_bench_48 | medium | true |
6be05541db3a4c8f8e82169353ed6aa2 | coding | null | [
"import (\n \"fmt\"\n)\n\n// Input are two strings a and b consisting only of 1s and 0s.\n// Perform binary XOR on these inputs and return result also as a string.\n// >>> StringXor('010', '110')\n// '100'\nfunc StringXor(a string, b string) string {\n"
] | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Go/11 | null | false |
94188a5a701c4394aabfb2214a6572b9 | coding | null | [
"FULL BENCHMARK DATA SHOULD BE FETCHED FROM THE SOURCE USING SPECDEC_BENCH\n\nYou will be given a question (problem specification) and will generate a correct program, in python programming language, that matches the specification and passes all tests.\n\nQuestion: {question}\n\nRead the inputs from stdin, solve th... | https://huggingface.co/datasets/livecodebench/code_generation_lite/tree/0fe84c3912ea0c4d4a78037083943e8f0c4dd505 | abc314_a | easy | false |
74f48ed284614abda4e67c724e4d6af7 | coding | null | [
"/*\nYou have to write a function which validates a given date string and\nreturns true if the date is valid otherwise false.\nThe date is valid if all of the following rules are satisfied:\n1. The date string is not empty.\n2. The number of days is not less than 1 or higher than 31 days for months 1,3,5,7,8,10,12.... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | CPP/124 | null | false |
4ca88202f05d4923b0fa2dcb3b6b637a | coding | null | [
"/*\n primeFib returns n-th number that is a Fibonacci number and it's also prime.\n >>> primeFib(1)\n 2\n >>> primeFib(2)\n 3\n >>> primeFib(3)\n 5\n >>> primeFib(4)\n 13\n >>> primeFib(5)\n 89\n */\nconst primeFib = (n) => {\n"
] | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | JavaScript/39 | null | false |
eaad9dbf75df4d2a90e328b155fcc978 | coding | null | [
"import java.util.*;\nimport java.lang.*;\n\nclass Solution {\n /**\n For a given string, flip lowercase characters to uppercase and uppercase to lowercase.\n >>> flipCase(\"Hello\")\n \"hELLO\"\n */\n public String flipCase(String string) {\n"
] | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Java/27 | null | false |
4ded8a39d0e64298848d673579638960 | coding | null | [
"FULL BENCHMARK DATA SHOULD BE FETCHED FROM THE SOURCE USING SPECDEC_BENCH\n\nYou will be given a question (problem specification) and will generate a correct program, in cpp programming language, that matches the specification and passes all tests.\n\nQuestion: {question}\n\nStarter code:\n ```\n{starter_code}\n``... | https://huggingface.co/datasets/livecodebench/code_generation_lite/tree/0fe84c3912ea0c4d4a78037083943e8f0c4dd505 | 2857 | easy | false |
126235195ce847b09b1dab1507421e64 | coding | null | [
"\ndef is_nested(string):\n '''\n Create a function that takes a string as input which contains only square brackets.\n The function should return True if and only if there is a valid subsequence of brackets \n where at least one bracket in the subsequence is nested.\n\n is_nested('[[]]') ➞ True\n ... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Python/132 | null | false |
73c564f6e702456fa85e903de4824d3c | coding | null | [
"Implement a program to find the common elements in two arrays without using any extra data structures.",
"Now the constraint of not using extra data structure is removed, implement one with the best time complexity."
] | https://raw.githubusercontent.com/hemingkx/Spec-Bench/66230f10cb0a02aced5ef3ce1e85163c16160454/data/spec_bench/question.jsonl | spec_bench_49 | medium | true |
5de62efe068247c4bb29dd5b37b21d80 | coding | null | [
"import (\n \"strings\"\n)\n\n// Create a function Encrypt that takes a string as an argument and\n// returns a string Encrypted with the alphabet being rotated.\n// The alphabet should be rotated in a manner such that the letters\n// shift down by two multiplied to two places.\n// For example:\n// Encrypt('hi')... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Go/89 | null | false |
2afb354c2f2d428e84a21770ef617b4f | coding | null | [
"/*\nYou will be given the name of a class (a string) and a vector of extensions.\nThe extensions are to be used to load additional classes to the class. The\nstrength of the extension is as follows: Let CAP be the number of the uppercase\nletters in the extension's name, and let SM be the number of lowercase lette... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | CPP/153 | null | false |
8fbbe76d4116423dae03ecd9e61b90d5 | coding | null | [
"FULL BENCHMARK DATA SHOULD BE FETCHED FROM THE SOURCE USING SPECDEC_BENCH\n\nYou will be given a problem specification and will generate a correct python program, that matches the specification and passes all tests.\n\n{question}\n\n"
] | https://huggingface.co/datasets/deepmind/code_contests/tree/802411c3010cb00d1b05bad57ca77365a3c699d6 | 1580_C. Train Maintenance | easy | false |
d21f707273ae4e16b80365e6c5394df4 | coding | null | [
"/* Create a function that returns true if the last character\n of a given string is an alphabetical character and is not\n a part of a word, and false otherwise.\n Note: \"word\" is a group of characters separated by space.\n Examples:\n checkIfLastCharIsALetter(\"apple pie\") ➞ false\n checkIfLastCharIsALet... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | JavaScript/134 | null | false |
84924e6ca98b4dfda5c32c335dc24324 | coding | null | [
"\n// There are eight planets in our solar system: the closerst to the Sun\n// is Mercury, the next one is Venus, then Earth, Mars, Jupiter, Saturn,\n// Uranus, Neptune.\n// Write a function that takes two planet names as strings planet1 and planet2.\n// The function should return a tuple containing all planets who... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Go/148 | null | false |
a1fa94f6735545c4821bec31321ab404 | coding | null | [
"\ndef is_multiply_prime(a):\n \"\"\"Write a function that returns true if the given number is the multiplication of 3 prime numbers\n and false otherwise.\n Knowing that (a) is less then 100. \n Example:\n is_multiply_prime(30) == True\n 30 = 2 * 3 * 5\n \"\"\"\n"
] | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Python/75 | null | false |
0a7a83477d19422db773738a353b2795 | coding | null | [
"import java.util.*;\nimport java.lang.*;\n\nclass Solution {\n /**\n For a given list of input numbers, calculate Mean Absolute Deviation\n around the mean of this dataset.\n Mean Absolute Deviation is the average absolute difference between each\n element and a centerpoint (mean in this case):\n ... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Java/4 | null | false |
68a3c3535f404350a2ece3e8c62211b5 | coding | null | [
"fn main(){}\n\nuse std::{slice::Iter, cmp::{max, self}, mem::replace, collections::{HashSet, HashMap}, ops::Index, ascii::AsciiExt};\nuse rand::Rng;\nuse regex::Regex;\nuse md5;\nuse std::any::{Any, TypeId};\n\n/*\n\n Given a dictionary, return True if all keys are strings in lower \n case or all keys are st... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Rust/95 | null | false |
9bb734657ad7475589f27e94941aad68 | coding | null | [
"/*\nYou will be given a string of words separated by commas or spaces. Your task is\nto split the string into words and return a vector of the words.\n\nFor example:\nwords_string(\"Hi, my name is John\") == {\"Hi\", \"my\", \"name\", \"is\", \"John\"}\nwords_string(\"One, two, three, four, five, six\") == {\"One\... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | CPP/101 | null | false |
278ebab0c4934903a93d6eb1e213f6cc | coding | null | [
" \n// Implement the Function F that takes n as a parameter,\n// and returns a list of size n, such that the value of the element at index i is the factorial of i if i is even\n// or the sum of numbers from 1 to i otherwise.\n// i starts from 1.\n// the factorial of i is the multiplication of the numbers from 1 to ... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Go/106 | null | false |
230db901c3d94ac4adfc7716753ea7e5 | coding | null | [
"FULL BENCHMARK DATA SHOULD BE FETCHED FROM THE SOURCE USING SPECDEC_BENCH\n\nYou will be given a question (problem specification) and will generate a correct program, in java programming language, that matches the specification and passes all tests.\n\nQuestion: {question}\n\nRead the inputs from stdin, solve the ... | https://huggingface.co/datasets/livecodebench/code_generation_lite/tree/0fe84c3912ea0c4d4a78037083943e8f0c4dd505 | abc305_a | easy | false |
326e163b8bc24f74abf125254d60051b | coding | null | [
"\ndef by_length(arr):\n \"\"\"\n Given an array of integers, sort the integers that are between 1 and 9 inclusive,\n reverse the resulting array, and then replace each digit by its corresponding name from\n \"One\", \"Two\", \"Three\", \"Four\", \"Five\", \"Six\", \"Seven\", \"Eight\", \"Nine\".\n\n ... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Python/105 | null | false |
c3a8af3647d342c9b7bf92ce6c550b26 | coding | null | [
"import java.util.*;\nimport java.lang.*;\n\nclass Solution {\n /**\n Write a function that returns True if the object q will fly, and False otherwise.\n The object q will fly if it's balanced (it is a palindromic list) and the sum of its elements is less than or equal the maximum possible weight w.\n\n ... | https://huggingface.co/datasets/bigcode/humanevalpack/tree/9a41762f73a8cb23bb5811b73d5aab164efcf378 | Java/72 | null | false |
69d831e14137427fa31d6e67761555e7 | humanities | null | [
"Provide insights into the correlation between economic indicators such as GDP, inflation, and unemployment rates. Explain how fiscal and monetary policies affect those indicators.",
"Now, explain them again like I'm five."
] | https://raw.githubusercontent.com/hemingkx/Spec-Bench/66230f10cb0a02aced5ef3ce1e85163c16160454/data/spec_bench/question.jsonl | spec_bench_70 | medium | true |
b8832921930f4ddd835e982a25655aad | humanities | History | [
"FULL BENCHMARK DATA SHOULD BE FETCHED FROM THE SOURCE USING SPECDEC_BENCH"
] | https://huggingface.co/datasets/cais/hle/tree/021a3d71f516a7ac28ceb8d284969902edf1edeb | 6736cec9cad1584306eee714 | hard | false |
90443bcccb5245b99cf1bd881fd07b33 | humanities | Archaeology | [
"FULL BENCHMARK DATA SHOULD BE FETCHED FROM THE SOURCE USING SPECDEC_BENCH"
] | https://huggingface.co/datasets/cais/hle/tree/021a3d71f516a7ac28ceb8d284969902edf1edeb | 671590d09f939ab157188104 | hard | false |
8884100957344de0a9304c38f8303f1d | humanities | Comparative Literature | [
"FULL BENCHMARK DATA SHOULD BE FETCHED FROM THE SOURCE USING SPECDEC_BENCH"
] | https://huggingface.co/datasets/cais/hle/tree/021a3d71f516a7ac28ceb8d284969902edf1edeb | 672fb1872357e1d1acd486dc | hard | false |
622f7fa890e14cc081e63bc90170c4cc | humanities | Corporate Eco-Efficiency | [
"FULL BENCHMARK DATA SHOULD BE FETCHED FROM THE SOURCE USING SPECDEC_BENCH"
] | https://huggingface.co/datasets/cais/hle/tree/021a3d71f516a7ac28ceb8d284969902edf1edeb | 6722072ddcce66512a82d9ed | hard | false |
5359bceeb76b4ec89e7f0f5cdd0eef21 | humanities | Classics | [
"FULL BENCHMARK DATA SHOULD BE FETCHED FROM THE SOURCE USING SPECDEC_BENCH"
] | https://huggingface.co/datasets/cais/hle/tree/021a3d71f516a7ac28ceb8d284969902edf1edeb | 6725703471b3b52354976be1 | hard | false |
6023d8b3dd4e431f884661d68a080e0d | humanities | Philosophy | [
"FULL BENCHMARK DATA SHOULD BE FETCHED FROM THE SOURCE USING SPECDEC_BENCH"
] | https://huggingface.co/datasets/cais/hle/tree/021a3d71f516a7ac28ceb8d284969902edf1edeb | 66fe796bc12cbe6aeaa26e69 | hard | false |
4ecf0309246b44bfa5cb3f923ada6909 | humanities | Linguistics | [
"FULL BENCHMARK DATA SHOULD BE FETCHED FROM THE SOURCE USING SPECDEC_BENCH"
] | https://huggingface.co/datasets/cais/hle/tree/021a3d71f516a7ac28ceb8d284969902edf1edeb | 66f2e7d9384d43333482ba75 | hard | false |
34bfdceb252d4c9d9bf7292918d0b009 | humanities | Finance | [
"FULL BENCHMARK DATA SHOULD BE FETCHED FROM THE SOURCE USING SPECDEC_BENCH"
] | https://huggingface.co/datasets/cais/hle/tree/021a3d71f516a7ac28ceb8d284969902edf1edeb | 66f248f6a9be6bc83b5b57f7 | hard | false |
dbcd535e14ac434ab14febcb9fd20591 | humanities | History | [
"FULL BENCHMARK DATA SHOULD BE FETCHED FROM THE SOURCE USING SPECDEC_BENCH"
] | https://huggingface.co/datasets/cais/hle/tree/021a3d71f516a7ac28ceb8d284969902edf1edeb | 6725145d97743d26179494d6 | hard | false |
3144677d7cfa40bfac5f79706e0d89ad | humanities | History | [
"FULL BENCHMARK DATA SHOULD BE FETCHED FROM THE SOURCE USING SPECDEC_BENCH"
] | https://huggingface.co/datasets/cais/hle/tree/021a3d71f516a7ac28ceb8d284969902edf1edeb | 6708c1dc2f33698babc6479d | hard | false |
008581155fe44bb3a8c8ee0fc487accc | humanities | Cultural Studies | [
"FULL BENCHMARK DATA SHOULD BE FETCHED FROM THE SOURCE USING SPECDEC_BENCH"
] | https://huggingface.co/datasets/cais/hle/tree/021a3d71f516a7ac28ceb8d284969902edf1edeb | 6710229bd8d5eea22faea867 | hard | false |
69b200ed82ad4a4a94d79f71a2e55f60 | humanities | Religious Studies, Classics, Trivia | [
"FULL BENCHMARK DATA SHOULD BE FETCHED FROM THE SOURCE USING SPECDEC_BENCH"
] | https://huggingface.co/datasets/cais/hle/tree/021a3d71f516a7ac28ceb8d284969902edf1edeb | 672d4dbdf618de6356b2fa89 | hard | false |
e36c236a0f9e4680b0c912df870c1337 | humanities | Foremost Classical: Period Functions | [
"FULL BENCHMARK DATA SHOULD BE FETCHED FROM THE SOURCE USING SPECDEC_BENCH"
] | https://huggingface.co/datasets/cais/hle/tree/021a3d71f516a7ac28ceb8d284969902edf1edeb | 672bb9be168a1a30f1f92fcc | hard | false |
c1cadb71190743d6aa36002687b6a652 | humanities | null | [
"What are some business etiquette norms when doing business in Japan?",
"Create a video script for training new employees of a car wash business in Japan. Highlight the above etiquette norms."
] | https://raw.githubusercontent.com/hemingkx/Spec-Bench/66230f10cb0a02aced5ef3ce1e85163c16160454/data/spec_bench/question.jsonl | spec_bench_78 | medium | true |
7e9b625014474ddb9802f49d49d3c241 | humanities | Linguistics | [
"FULL BENCHMARK DATA SHOULD BE FETCHED FROM THE SOURCE USING SPECDEC_BENCH"
] | https://huggingface.co/datasets/cais/hle/tree/021a3d71f516a7ac28ceb8d284969902edf1edeb | 673761e610c7e233967eef47 | hard | false |
af288f0c78e44a44956c83b24e735a1c | humanities | History | [
"FULL BENCHMARK DATA SHOULD BE FETCHED FROM THE SOURCE USING SPECDEC_BENCH"
] | https://huggingface.co/datasets/cais/hle/tree/021a3d71f516a7ac28ceb8d284969902edf1edeb | 66fc8f1b8a73f859d8fa1608 | hard | false |
9708caa32e7745be8e02d9a60230070a | humanities | Russian Literature | [
"FULL BENCHMARK DATA SHOULD BE FETCHED FROM THE SOURCE USING SPECDEC_BENCH"
] | https://huggingface.co/datasets/cais/hle/tree/021a3d71f516a7ac28ceb8d284969902edf1edeb | 6735c6e5f8dc0a94c3a3ae8e | hard | false |
5491c23305b24c228345a1851f62abe6 | humanities | Archaeology | [
"FULL BENCHMARK DATA SHOULD BE FETCHED FROM THE SOURCE USING SPECDEC_BENCH"
] | https://huggingface.co/datasets/cais/hle/tree/021a3d71f516a7ac28ceb8d284969902edf1edeb | 672365aa71fe0b446a1eb0a2 | hard | false |
7f06f76a0d9b44619dc88333ecf84804 | humanities | null | [
"Explain what's base rate fallacy and list five specific examples of how politicians use it for campaigns.",
"Provide a detailed plan for an election campaign using the first example."
] | https://raw.githubusercontent.com/hemingkx/Spec-Bench/66230f10cb0a02aced5ef3ce1e85163c16160454/data/spec_bench/question.jsonl | spec_bench_75 | medium | true |
- データの内容
- データの概要
- 複製について
- 出典
- 論文
- アノテーションの方法
- 微調整に使うときの注意
- 調査メモの記述
- 元のカード(日本語訳)
- データセットの詳細
- データセットの構造
- データセットの作成
- SpecBench との詳細なデータ比較
- データセットの所有者:
- データセットの作成日:
- ライセンス/利用規約:
- 想定される用途:
- データ形式
- データ量
- 参考文献:
- 倫理上の配慮:
- データセットカードの連絡先
- Dataset Details
- Dataset Structure
- Dataset Creation
- Detailed Data Comparison with SpecBench
- Dataset Owner(s):
- Dataset Creation Date:
- License/Terms of Use:
- Intended Usage:
- Dataset Format
- Dataset Quantification
- Reference(s):
- Ethical Considerations:
- Dataset Card Contact
nvidia/SPEED-Bench(複製)
データの内容
投機的復号(speculative decoding)の受理率とスループットを測るための評価用ベンチマークのデータ。2 種類の分割からなる。Qualitative は意味の多様さを優先して分野ごとに集めた問い、Throughput は入力の長さと、低・混合・高エントロピーの区分を変えた入力である。1 行が 1 件の問いで、主な列は question_id、category、turns。一部のサンプルは本文が伏せられており、付属の計測フレームワークでソースから取り直す必要がある。
データの概要
| 項目 | 値 |
|---|---|
| 行数 | 8,944(7 構成・7 分割の合計) |
| 列数 | 8 |
| 容量 | 141.59 MB(ファイル 15 本) |
| ファイルの形式 | parquet 7、png 5、pdf 1 |
| 公開 | public |
構成と分割
| 構成(config) | 分割(split)ごとの行数 | 列数 |
|---|---|---|
qualitative |
test 880 |
8 |
throughput_100k |
test 384 |
8 |
throughput_16k |
test 1,536 |
8 |
throughput_1k |
test 1,536 |
8 |
throughput_2k |
test 1,536 |
8 |
throughput_32k |
test 1,536 |
8 |
throughput_8k |
test 1,536 |
8 |
列(構成 qualitative・分割 test。どの構成も同じ列)
| 列 | 型 | 先頭の行の値 |
|---|---|---|
question_id |
string |
0daf539b787c4dccbb547330a8b4c3d7 |
category |
string |
coding |
sub_category |
string |
null |
turns |
list<string> |
[1 件] |
source |
string |
https://huggingface.co/datasets/bigcode/humaneva… |
src_id |
string |
Go/27 |
difficulty |
string |
null |
multiturn |
bool |
False |
行数と列は HF のデータセットビューアの集計(元のデータセットの現在の版で数えたもの)。全データセットの一覧は HYUKI5/dataset-catalog にある。
複製について
nvidia/SPEED-Bench をリビジョン fa4e669bee2a4393c56aa8ed05a7c667897f9590 の時点でそのまま複製したもの。中身は変えていない。判断モデル Laya(convaiinnovations/laya)の微調整に使う候補として、調査メモ(Laya 学習用データセット調査)に載せたデータを集めている。
出典
- 元のデータセット: https://huggingface.co/datasets/nvidia/SPEED-Bench
- 複製したリビジョン:
fa4e669bee2a4393c56aa8ed05a7c667897f9590(2026-11-04 12:54:26+00:00) - 元のライセンス表記:
other
論文
- https://arxiv.org/abs/2604.09557
- https://arxiv.org/abs/2401.07851
- https://openreview.net/forum?id=Rl2uQlCoQX
アノテーションの方法
キュレーションの根拠 - Qualitative split
Qualitative split の目的は、幅広い意味領域にわたって SD の品質、具体的には条件付きの受理率(AR)と受理長(AL)を測ることである。
多数のデータセットにわたる網羅的な評価は理論上は可能だが、手間がかかり、迅速な実験には非現実的で、SD のアルゴリズムやモデルを公開する異なる研究グループ同士の直接比較も妨げる。異なるデータセットにまたがる網羅的な評価に頼る代わりに、意味的な多様性を最大化するように設計された、小さいがよく代表する部分集合をキュレーションした。 18 の公開されたソースからデータを集め、コーディング、数学、人文学、STEM、執筆、要約、ロールプレイ、RAG、多言語、推論、QA を含む 11 のカテゴリに整理した。
カテゴリ内の多様性が低いことが多い従来のベンチマークと異なり、SPEED-Bench の Qualitative split は意味的な多様性を明示的に優先する。
これを実現するため、候補のプロンプトはそれぞれ、事前学習済みのテキスト埋め込みモデル(openai/text-embedding-3-small)を使って密なベクトル空間に埋め込む。
次に、カテゴリ内の組ごとのコサイン類似度の平均を最小化する選択アルゴリズムを適用する。
これにより、選ばれたサンプルが意味空間をできる限り広く覆い、冗長性を減らし、評価の忠実度を高める。
キュレーションの根拠 - Throughput split
Qualitative split は、ドラフターの正確さを捉えるが、システム全体の高速化を評価するには不十分である。 Throughput split は、まさにこれを捉えるために設計されている。
コーディング支援や retrieval-augmented generation のような長文脈のアプリケーションの重要性が高まっていることを反映して、1k から 32k token までの固定の ISL のバケットを作る。 各 ISL のバケットでは、プロンプトを低エントロピー、混合エントロピー、高エントロピーの領域に対応する 3 つの粗い難易度カテゴリにまとめる。 prefill のコストを決定的にするため、意味内容を保ったまま、プロンプトを制御された方法で切り詰めるかパディングする。
各 ISL のバケットには 1,536 のプロンプト(難易度カテゴリごとに 512)が含まれ、幅広いバッチサイズにわたって安定した throughput のパレート曲線を描くのに十分な量がある。
重要な点として、SPEED-Bench はスループットのベンチマークにランダムな token の入力を使わない。ランダムな token は、受理の振る舞い、MoE モデルのエキスパートのルーティング、スループットの計測を大きく歪め、楽観的すぎる結論を導きかねない。
データソース
このベンチマークでは次のデータソースを使う:
| Data Source | Categories | Construction Details |
|---|---|---|
| SpecBench | STEM、人文学、執筆、コーディング、数学、ロールプレイ、推論、QA、RAG | ソースからそのまま使用 |
| CNN/Daily Mail | 要約 | ソースの記事をそのまま使い、内容を要約する指示を付けた。 |
| Humanity's Last Exam | STEM、人文学、数学 | テキストのみのサンプル(画像なし)に絞り、答えの型が exact-match のものを残した。STEM は物理学、CS/AI、生物学/医学、化学、工学に絞った。人文学は Humanities/Social Science のカテゴリに絞った。 |
| LiveCodeBench Lite | コーディング | ランダムに選んだプログラミング言語(Python、Java、C++、Go、JavaScript、Rust)でのコード生成を求める指示のプロンプトを作った。利用できる場合は開始用のコードを含める。 |
| Code Contests | コーディング | ランダムに選んだ言語(Python、Java、C++)でのプログラム生成を求める指示のプロンプトを作った。問題文はソースからそのまま使用。 |
| HumanEvalPack | コーディング | コードの補完のプロンプトをソースからそのまま使用。 |
| RoleBench | ロールプレイ | 役の説明と質問を使い、複数ターンのロールプレイのプロンプトを作った。質問は役ごとにまとめて会話にした(1--5 ターン)。システムプロンプトは、モデルにキャラクターを演じさせる 8 種類のプロンプトのテンプレートからランダムに選んだ。 |
| CoSER | ロールプレイ | キャラクターの設定、シナリオ、キャラクターの動機を含むロールプレイのプロンプトを作った。対象はパブリックドメインで利用できる書籍のみ。 |
| WritingBench | 執筆 | 英語のサンプルに絞った。執筆の依頼文を、1 ターンのプロンプトとしてそのまま使用。 |
| Creative Writing V3 | 執筆 | <SEED> のプレースホルダを、提供されたシード修飾語で置き換えてプロンプトを展開し、基になるプロンプトごとに複数のバリエーションを作った。 |
| MT-Bench 101 | 推論 | 一般的な推論と数学的推論のタスクに絞った。 |
| MMLU-Pro | 推論 | 質問をカテゴリごとにまとめ、複数の質問を組み合わせて複数ターンのサンプルを作った。 |
| MMATH | 多言語 | 質問をソースからそのまま使用。 |
| OPUS-100 | 多言語 | プロンプトの先頭に "Translate the following text from [source language] to [target language]:" を付けて翻訳のプロンプトを作った。 |
| MCIF | 多言語 | QA、翻訳、要約のタスクのプロンプトを、long_mixed-prompt の形式で選んだ。 |
| ChatRAG-Bench | RAG | 文脈(取得した段落を連結したもの)と複数ターンの質問を含むプロンプトを、hybridial と sqa の split について作った。 |
| MCIF | RAG | 英語の QA のプロンプトを long_mixed-prompt の形式で使い、文書ごとに質問をまとめて複数ターンの会話にした。 |
| BAMBOO | Throughput split(執筆 - 高エントロピー) | MeetingPred と ShowsPred のサブセットを使った。会話の続きをモデルに書かせる対話の補完のプロンプトを作った。より長い文脈(>16k tokens)では、複数の対話を連結した。目標の token 数になるよう、パディングまたは切り詰めを行った。 |
| Project Gutenberg | Throughput split(執筆 - 高エントロピー) | 書籍の続きを書かせるプロンプトを作った。十分な長さのある書籍に絞り、目標の token 数になるようパディングまたは切り詰めを行った。 |
| WritingBench | Throughput split(執筆 - 高エントロピー) | Qualitative split の英語の執筆プロンプトを再利用した。目標の token 数の 0.7--2 倍の範囲のプロンプトに絞り、その後パディングまたは切り詰めを行った。 |
| AdaLEval (StackSelect) | Throughput split(NIAH - 混合) | 干し草の山の中の針(needle-in-a-haystack)のプロンプトを作った。StackOverflow の回答の集合から最も役に立つ回答を選び、各選択肢の説明を出すようモデルに求めた。目標の token 数になるようパディングまたは切り詰めを行った。 |
| Humanity's Last Exam | Throughput split(混合) | HLE のデータの 50% を few-shot のプロンプトに使った。保留しておいた例からサンプリングしたカテゴリごとのデモンストレーションを付け、その後に対象の質問を置いたプロンプトを作った。目標の token 数になるようパディングまたは切り詰めを行った。 |
| [Long Code Arena](https://huggingface.co/spaces/JetBrains-Research/long-code- |
(長いので途中で切った。全文は下の「元のカード」を参照)
微調整に使うときの注意
- 調査メモの台帳での区分: D(記載なし・other・unknown・資料間の食い違い)。区分は記載の文言で機械的に付けたもので、法的な判定ではない。学習に入れる前に元のカードとライセンスの原文で確かめる。
- Laya の入力は 2,048 token まで(ModernBERT-large / mmBERT)。長い本文は先頭から黙って切られるので、切れる行は事前に除くか、要約・切り出しをしてから入れる。
- 分割は記事・文書単位で行い、同じ本文から作った別の問いを学習側と較正側に分けない。温度と閾値は学習に使っていない分割で別々に当てる。
調査メモの記述
| メモ | 節 | ライセンス(記載) | 確認 | 注記 |
|---|---|---|---|---|
nvidia_nemotron.md |
付録. 取得した nvidia 組織のデータセット全件の簡易一覧 (L373) | other | 確認済 | 確認印なし(各ファイルの凡例により確認済扱い) / 原文注記: × 数学・コード・科学・計測 |
元のカード(日本語訳)
以下は元のデータセットのカードの本文を日本語に訳したもの(機械翻訳)。訳に疑問があるときは、末尾の原文を確かめる。
📒 ブログ | 📄 論文 | 🤗 データ | ⚙️ 計測フレームワーク
📰 ニュース
[2026年9月17日] v1.1 を公開した。リリースノート:
- 128 プロンプトずつのエントロピーのカテゴリ(低、混合、高)による 100K の throughput split を追加した。合計 384 プロンプト。
- 16K、32K、100K の throughput split の低エントロピーのカテゴリに、**InfiniteBench のコードのデバッグのタスク** を追加した。これらのプロンプトでは、タスクの指示をコードの文脈より前の先頭に置く。
- バグ修正:
- throughput のデータ分割における RepoBench のサンプリングを更新し、言語とタスクの設定ごとに、リポジトリ間で選択が偏らないようにして、異なる文脈を優先するようにした。
- throughput のデータ分割における HLE のサンプリングを更新し、プロンプト内のデモンストレーションが互いに異なることを必須にした。
- qualitative のデータ分割の、OPUS-100 由来の多言語のサンプルから、漏れていた正解の翻訳を取り除いた。
- MMLU-Pro 由来の qualitative のデータ分割の推論のサンプルについて、
multiturnフィールドを修正した。- MMLU-Pro 由来の qualitative のデータ分割の推論のサンプルについて、質問の分野を修正した。
SPEED-Bench(SPEculative Evaluation Dataset、投機的復号の評価用データセット)は、本番環境向けの推論エンジンを使い、多様な意味領域と現実的な提供の条件にわたって投機的復号(SD)を評価するために設計された統一ベンチマークである。 受理率の特性と、エンドツーエンドのスループットの両方を計測し、SD の手法同士を公平で再現可能かつ頑健に比較できるようにする。 SPEED-Bench は、SD のためのベンチマークのエコシステムを導入する。 2 つの目的別に作られたデータ分割と、統一された計測フレームワークを組み合わせている。それぞれが SD の振る舞いの異なる側面を捉えるように設計されている:
意味的な多様性に最適化され、領域をまたいで推測の品質(ドラフターの正確さ)を計測するように設計された 「Qualitative」の data split。
さまざまな入力系列長と高い同時実行数にわたって、システム全体の性能を評価するために構成された 「Throughput」の data split。
本番環境の推論エンジンと統合され、システムをまたいだ評価を標準化する 統一計測フレームワーク。
データセットの詳細
データセットの説明
SPEED-Bench は、SD のアルゴリズムの評価を標準化し、改善するために設計された統一ベンチマークである。
SpecBench は、広く使われているデータセットのインスタンスを統一されたテスト環境にまとめることで、多ターンの会話、翻訳、数学的推論などの多様な応用シナリオにわたる、初の統一された SD ベンチマークを導入した。しかし、標準化された評価に向けた重要な一歩である一方で、規模と多様性に関して重大な限界がある。ほとんどのカテゴリは 10 サンプルしかなく、平均の入力長も短い(100 token 未満)ため、最近のドラフターに負荷をかけられないことがある。加えて、一部のカテゴリは構造的な多様性を欠くことが多い。たとえば、多言語のカテゴリはすべてドイツ語から英語への翻訳のプロンプトからなる。
本ベンチマークは、2 つの主要なデータセットの構成を導入する:
Qualitative の data split: この構成は既存の複数のベンチマークを組み合わせて洗練したもので、元の SpecBench に似たカテゴリを扱う。18 の公開されたソースからデータを集め、コーディング、数学、人文学、STEM、執筆、要約、ロールプレイ、RAG、多言語、推論、QA を含む 11 のカテゴリに整理した。カテゴリ内の多様性が低いことが多い従来のベンチマークと異なり、SPEED-Bench の Qualitative split は意味的な多様性を明示的に優先する。異なるタスクの領域にわたる SD のアルゴリズムの受理率と総合的な性能を評価するように設計されている。
Throughput の data split: この構成は、固定長の入力系列(1K、2K、8K、16K、32K)からなり、3 つの粗い難易度カテゴリに整理されている。低エントロピーの領域(コーディングなど)、高エントロピーの領域(創作的な執筆など)、混合エントロピーの領域である。目的は、高い同時実行数と幅広い入力系列長(ISL)のもとで throughput を評価することであり、その長さは多くの SD ベンチマークで使われる短い ISL のサンプルよりずっと長いことが多い。バッチサイズが大きくなると、推論は計算律速の領域からメモリ律速の領域へ移ることが多く、投機的復号のコストと利益のトレードオフが根本的に変わる。
| ⚠️ 重要: データセットの一部のサンプルは、文 "FULL BENCHMARK DATA SHOULD BE FETCHED FROM THE SOURCE USING SPECDEC_BENCH" でマスクされている。完全なベンチマークのデータを得るには、付属の計測フレームワークを使う。これはソースから完全なデータを直接取得する。 |
|---|
データセットの構造
各データ点は次のフィールドを含む:
question_id- SPEED-Bench 内のデータ点の一意の識別子。category- データ点のカテゴリ- qualitative の場合は次のいずれか: coding, stem, humanities, math, multilingual, qa, rag, roleplay, reasoning, summarization, writing。
- throughput の場合は次のいずれか: high_entropy, mixed, low_entropy。
sub_category- 元のデータソースが提供している場合の、データ点の細かいカテゴリ。turns- 会話を組み立てるのに使うプロンプトのリスト。source- 元のデータソースの URL。src_id- データソースにおけるサンプルの元の ID。difficulty- 元のデータソースが提供している場合の、サンプルの難易度。multiturn- データ点が複数ターンの会話を含むかを示す真偽値のフィールド(つまりlen(turns) > 1)。
データセットの作成
キュレーションの根拠 - Qualitative split
Qualitative split の目的は、幅広い意味領域にわたって SD の品質、具体的には条件付きの受理率(AR)と受理長(AL)を測ることである。
多数のデータセットにわたる網羅的な評価は理論上は可能だが、手間がかかり、迅速な実験には非現実的で、SD のアルゴリズムやモデルを公開する異なる研究グループ同士の直接比較も妨げる。異なるデータセットにまたがる網羅的な評価に頼る代わりに、意味的な多様性を最大化するように設計された、小さいがよく代表する部分集合をキュレーションした。 18 の公開されたソースからデータを集め、コーディング、数学、人文学、STEM、執筆、要約、ロールプレイ、RAG、多言語、推論、QA を含む 11 のカテゴリに整理した。
カテゴリ内の多様性が低いことが多い従来のベンチマークと異なり、SPEED-Bench の Qualitative split は意味的な多様性を明示的に優先する。
これを実現するため、候補のプロンプトはそれぞれ、事前学習済みのテキスト埋め込みモデル(openai/text-embedding-3-small)を使って密なベクトル空間に埋め込む。
次に、カテゴリ内の組ごとのコサイン類似度の平均を最小化する選択アルゴリズムを適用する。
これにより、選ばれたサンプルが意味空間をできる限り広く覆い、冗長性を減らし、評価の忠実度を高める。
サンプリングのアルゴリズム
カテゴリごとに、意味的な多様性を最大化する 80 サンプルの部分集合を選ぶことを目標とする。 その動機は、多様な入力にわたって投機的復号のアルゴリズムの受理率をより信頼できる形で計測することである。
手順:
- 埋め込みの計算: OpenAI の
text-embedding-3-smallモデルを使い、カテゴリ内のすべてのサンプルの埋め込みを計算する。 - 貪欲選択: ランダムなサンプルから始め、既に選んだ部分集合との意味的な類似度を最小化するサンプルを、80 サンプルに達するまで反復的に選ぶ。
- 部分集合の最適化: 最初の選択の後、選んだサンプルと選ばなかったサンプルを入れ替え、類似度をさらに下げて意味的な多様性を高める。
次の表は、ランダム選択と(上のサンプリングのアルゴリズムで述べた)貪欲選択で選んだ SPEED-Bench の部分集合と比べた、SpecBench のサンプルの平均の意味的類似度を示す。
すべての値は、OpenAI の text-embedding-3-small モデルで生成した埋め込みに対するコサイン類似度で計算した。
値が小さいほど意味的な多様性が高いことを示す。括弧内の数値は SpecBench と比べた相対的な変化を示す。
| Category | SpecBench | SPEED (random selection) | SPEED (greedy algorithm) |
|---|---|---|---|
| 数学 | 0.24 | 0.21 (-12.5%) | 0.15 (-37.5%) |
| コーディング | 0.33 | 0.48 (+45%) | 0.16 (-51%) |
| 多言語 | 0.36 | 0.18 (-50%) | 0.06 (-83%) |
| 執筆 | 0.35 | 0.29 (-17%) | 0.18 (-48%) |
| 人文学 | 0.26 | 0.14 (-46%) | 0.12 (-54%) |
| STEM | 0.24 | 0.19 (-21%) | 0.12 (-50%) |
| 推論 | 0.24 | 0.21 (-12.5%) | 0.14 (-41%) |
| RAG | 0.15 | 0.17 (+13%) | 0.13 (-13%) |
| ロールプレイ | 0.28 | 0.25 (-11%) | 0.24 (-14%) |
| 要約 | 0.21 | 0.24 (+14%) | 0.11 (-48%) |
| QA | 0.10 | 変更されていないサンプル | 変更されていないサンプル |
キュレーションの根拠 - Throughput split
Qualitative split は、ドラフターの正確さを捉えるが、システム全体の高速化を評価するには不十分である。 Throughput split は、まさにこれを捉えるために設計されている。
コーディング支援や retrieval-augmented generation のような長文脈のアプリケーションの重要性が高まっていることを反映して、1k から 32k token までの固定の ISL のバケットを作る。 各 ISL のバケットでは、プロンプトを低エントロピー、混合エントロピー、高エントロピーの領域に対応する 3 つの粗い難易度カテゴリにまとめる。 prefill のコストを決定的にするため、意味内容を保ったまま、プロンプトを制御された方法で切り詰めるかパディングする。
各 ISL のバケットには 1,536 のプロンプト(難易度カテゴリごとに 512)が含まれ、幅広いバッチサイズにわたって安定した throughput のパレート曲線を描くのに十分な量がある。
重要な点として、SPEED-Bench はスループットのベンチマークにランダムな token の入力を使わない。ランダムな token は、受理の振る舞い、MoE モデルのエキスパートのルーティング、スループットの計測を大きく歪め、楽観的すぎる結論を導きかねない。
データソース
このベンチマークでは次のデータソースを使う:
| Data Source | Categories | Construction Details |
|---|---|---|
| SpecBench | STEM、人文学、執筆、コーディング、数学、ロールプレイ、推論、QA、RAG | ソースからそのまま使用 |
| CNN/Daily Mail | 要約 | ソースの記事をそのまま使い、内容を要約する指示を付けた。 |
| Humanity's Last Exam | STEM、人文学、数学 | テキストのみのサンプル(画像なし)に絞り、答えの型が exact-match のものを残した。STEM は物理学、CS/AI、生物学/医学、化学、工学に絞った。人文学は Humanities/Social Science のカテゴリに絞った。 |
| LiveCodeBench Lite | コーディング | ランダムに選んだプログラミング言語(Python、Java、C++、Go、JavaScript、Rust)でのコード生成を求める指示のプロンプトを作った。利用できる場合は開始用のコードを含める。 |
| Code Contests | コーディング | ランダムに選んだ言語(Python、Java、C++)でのプログラム生成を求める指示のプロンプトを作った。問題文はソースからそのまま使用。 |
| HumanEvalPack | コーディング | コードの補完のプロンプトをソースからそのまま使用。 |
| RoleBench | ロールプレイ | 役の説明と質問を使い、複数ターンのロールプレイのプロンプトを作った。質問は役ごとにまとめて会話にした(1--5 ターン)。システムプロンプトは、モデルにキャラクターを演じさせる 8 種類のプロンプトのテンプレートからランダムに選んだ。 |
| CoSER | ロールプレイ | キャラクターの設定、シナリオ、キャラクターの動機を含むロールプレイのプロンプトを作った。対象はパブリックドメインで利用できる書籍のみ。 |
| WritingBench | 執筆 | 英語のサンプルに絞った。執筆の依頼文を、1 ターンのプロンプトとしてそのまま使用。 |
| Creative Writing V3 | 執筆 | <SEED> のプレースホルダを、提供されたシード修飾語で置き換えてプロンプトを展開し、基になるプロンプトごとに複数のバリエーションを作った。 |
| MT-Bench 101 | 推論 | 一般的な推論と数学的推論のタスクに絞った。 |
| MMLU-Pro | 推論 | 質問をカテゴリごとにまとめ、複数の質問を組み合わせて複数ターンのサンプルを作った。 |
| MMATH | 多言語 | 質問をソースからそのまま使用。 |
| OPUS-100 | 多言語 | プロンプトの先頭に "Translate the following text from [source language] to [target language]:" を付けて翻訳のプロンプトを作った。 |
| MCIF | 多言語 | QA、翻訳、要約のタスクのプロンプトを、long_mixed-prompt の形式で選んだ。 |
| ChatRAG-Bench | RAG | 文脈(取得した段落を連結したもの)と複数ターンの質問を含むプロンプトを、hybridial と sqa の split について作った。 |
| MCIF | RAG | 英語の QA のプロンプトを long_mixed-prompt の形式で使い、文書ごとに質問をまとめて複数ターンの会話にした。 |
| BAMBOO | Throughput split(執筆 - 高エントロピー) | MeetingPred と ShowsPred のサブセットを使った。会話の続きをモデルに書かせる対話の補完のプロンプトを作った。より長い文脈(>16k tokens)では、複数の対話を連結した。目標の token 数になるよう、パディングまたは切り詰めを行った。 |
| Project Gutenberg | Throughput split(執筆 - 高エントロピー) | 書籍の続きを書かせるプロンプトを作った。十分な長さのある書籍に絞り、目標の token 数になるようパディングまたは切り詰めを行った。 |
| WritingBench | Throughput split(執筆 - 高エントロピー) | Qualitative split の英語の執筆プロンプトを再利用した。目標の token 数の 0.7--2 倍の範囲のプロンプトに絞り、その後パディングまたは切り詰めを行った。 |
| AdaLEval (StackSelect) | Throughput split(NIAH - 混合) | 干し草の山の中の針(needle-in-a-haystack)のプロンプトを作った。StackOverflow の回答の集合から最も役に立つ回答を選び、各選択肢の説明を出すようモデルに求めた。目標の token 数になるようパディングまたは切り詰めを行った。 |
| Humanity's Last Exam | Throughput split(混合) | HLE のデータの 50% を few-shot のプロンプトに使った。保留しておいた例からサンプリングしたカテゴリごとのデモンストレーションを付け、その後に対象の質問を置いたプロンプトを作った。目標の token 数になるようパディングまたは切り詰めを行った。 |
| Long Code Arena | Throughput split(コーディング - 低エントロピー) | プロジェクト単位のコード補完のサブセットを使った。リポジトリの文脈と、行単位の補完のための [COMPLETE] マーカーを含むファイルでプロンプトを作った。 |
| RepoBench Python | Throughput split(コーディング - 低エントロピー) | リポジトリの文脈の断片とファイル内のコードを含む、ファイル横断のコード補完のプロンプトを作った。目標の token 数になるようパディングまたは切り詰めを行った。 |
| RepoBench Java | Throughput split(コーディング - 低エントロピー) | 作り方は RepoBench Python と同じで、Java のコード向け。 |
| AdaLEval (TextSort) | Throughput split(並べ替え - 低エントロピー) | 元の並べ替えのタスクを、インデックスだけを返すのではなく、並べ替えたテキストの断片を順番どおりに出力するよう変更した。目標の token 数になるようパディングまたは切り詰めを行った。 |
SpecBench との詳細なデータ比較
| Metric | SPEED-Bench | SpecBench |
|---|---|---|
| # カテゴリごとのサンプル数 | 80 (qualitative)、512 x 3 (throughput) | 10(8 カテゴリ)、80(残り) |
| # サンプル総数 | 880 (qualitative)、1536 x 5 (throughput) | 480 |
| # データソース数 | 24 | 5 |
| 平均の組ごとの類似度 | 0.14 | 0.22 |
| # 複数ターンのプロンプト数 | 167 | 80 |
| 最大ターン数 | 5 | 2 |
| サブカテゴリ | ✅ | ❌ |
| 難易度 | ✅(数学、STEM、人文学、コーディング について) | ❌ |
| 長い ISL(16k-32k) | ✅ | ❌ |
| 固定長の ISL の大きなバッチ | ✅ | ❌ |
| コーディング で明示的に言及されたプログラミング言語 | Python (27), CPP (9), Java (10), Go (13), Javascript (11), Rust (3), HTML (1), CSS (1) | Python (3), CPP (1), HTML (1), CSS (1) |
| 多言語 の異なる言語数 | 23 | 2 |
| 多言語 の言語 | EN, DE, ZH, IT, MG, FR, JA, PT, AR, MK, DA, NL, KO, ES, NN, TH, VI, BN, GU, CS, GD, EU, RU | EN, DE |
| 数学、人文学、STEM カテゴリの難易度 | 学術レベル | 高校レベル |
2 つのカテゴリについての SpecBench と SPEED-Bench の類似度行列の例(値が小さいほど良い):
データセットの所有者:
NVIDIA Corporation
データセットの作成日:
2026年2月
ライセンス/利用規約:
適用される規約: このデータセットは NVIDIA Evaluation Dataset License Agreement に従う。
補足情報: bigcode/humanevalpack、RUCAIBox/MMATH、RUCAIBox/BAMBOO、EQ-Bench は MIT。Writing Bench と Spec-Bench は Apache 2.0。FBK-MT/MCIF は CC BY 4.0。tianyang/repobench_python_v1.1、JetBrains-Research/lca-project-level-code-completion、tianyang/repobench_java_v1.1 は MIT と Apache 2.0。
注意: nvidia/Model-Optimizer GitHub リポジトリ にある計測フレームワークに含まれる prepare_data.py スクリプトは、すべてのソースのデータセットからデータを自動的に取得し、組み立てる。利用者は使うデータセットごとに、そのライセンスが目的に合うかを自ら確認する責任を負う。
想定される用途:
このデータセットは、投機的復号(Speculative Decoding、SD)と LLM 推論のレイテンシ計測を評価する LLM エンジニアと研究者向けに設計されている。
データ形式
データセットは parquet 形式で配布される。ライセンスの遵守を確実にするため、ハイブリッドな配布の仕組みを取る。 配布の構造:
- サンプルの一部は parquet ファイルに直接格納されている。
- サンプルの一部はプレースホルダの注記と、データソースへの参照を含む。元のホストの場所から直接データを取得するデータセット構築スクリプトを提供している。このスクリプトを使うことで、利用者は第三者からデータをダウンロードしていることを認め、それぞれのライセンス条件を遵守する責任を負うものとする。
データ量
- レコード数: qualitative split は 880 サンプル、throughput split は 1538 サンプル。
- データの総容量: 約 95MB
参考文献:
論文: https://arxiv.org/abs/2604.09557 .
SPEED-Bench が役に立ったなら、以下の引用を検討してほしい。
@inproceedings{
abramovich2026speedbench,
title={{SPEED}-Bench: A Unified and Diverse Benchmark for Speculative Decoding},
author={Talor Abramovich and Maor Ashkenazi and Izzy Putterman and Benjamin Chislett and Tiyasa Mitra and Bita Darvish Rouhani and Ran Zilberstein and Yonatan Geifman},
booktitle={Forty-third International Conference on Machine Learning},
year={2026},
url={https://openreview.net/forum?id=Rl2uQlCoQX}
}
倫理上の配慮:
NVIDIA は、信頼できる AI は共有の責任であると考えており、幅広い AI アプリケーションの開発を可能にするための方針と実践を定めている。利用規約に従ってダウンロードまたは使用する場合、開発者は社内のチームと協力して、このデータセットが関連する業界と用途の要件を満たし、想定外の製品の悪用に対処しているかを確かめるべきである。 品質、リスク、セキュリティの脆弱性、または NVIDIA の AI に関する懸念は、こちらから報告してほしい。
データセットカードの連絡先
Talor Abramovich - talora@nvidia.com
Maor Ashkenazi - mashkenazi@nvidia.com
原文(英語)
📒 Blog | 📄 Paper | 🤗 Data | ⚙️ Measurement Framework
📰 News
[September 17, 2026] Released v1.1. Release notes:
- Added a 100K throughput split with 128 prompts per entropy category (low, mixed, and high), for 384 prompts total.
- Added InfiniteBench code-debugging tasks to the low-entropy category of the 16K, 32K, and 100K throughput splits. These prompts place the task instructions at the beginning, before the code context.
- Bug fixes:
- Updated RepoBench sampling in the throughput data split to balance selection across repositories within each language and task setting to prioritize distinct contexts.
- Updated HLE sampling in the throughput data split to require distinct demonstration in the prompt.
- Removed leaked target translations from qualitative data split multilingual examples sourced from OPUS-100.
- Corrected the
multiturnfield for qualitative data split reasoning examples sourced from MMLU-Pro.- Corrected the questions subjects for qualitative data split reasoning examples sourced from MMLU-Pro.
SPEED-Bench (SPEculative Evaluation Dataset) is a unified benchmark designed to evaluate speculative decoding (SD) across diverse semantic domains and realistic serving regimes, using production-grade inference engines. It measures both acceptance-rate characteristics and end-to-end throughput, enabling fair, reproducible, and robust comparisons between SD strategies. SPEED-Bench introduces a benchmarking ecosystem for SD. It combines two purpose-built dataset splits and a unified measurement framework, each designed to capture a different aspect of SD behavior:
A "Qualitative" data split, optimized for semantic diversity and designed to measure speculation quality (drafter accuracy) across domains.
A "Throughput" data split, constructed to evaluate system-level performance across various input sequence lengths and high concurrency.
A unified measurement framework, integrated with production inference engines, that standardizes evaluation across systems.
Dataset Details
Dataset Description
SPEED-Bench is a unified benchmark designed to standardize and improve the evaluation of SD algorithms.
SpecBench introduced the first unified SD benchmark across diverse application scenarios, such as multi-turn conversation, translation, and mathematical reasoning, by aggregating instances from widely used datasets into a unified testing environment. However, despite being a significant step toward standardized evaluations, it has critical limitations regarding scale and diversity. Most categories contain as few as 10 samples with short mean input lengths (< 100 tokens) that may fail to stress modern drafters. Additionally, some of its categories often lack structural diversity, such as the multilingual category consisting entirely of German-to-English translation prompts.
Our benchmark introduces two main dataset configurations:
A Qualitative data split: This configuration combines and refines multiple existing benchmarks, covering categories similar to the original SpecBench. We aggregate data from 18 publicly available sources and organize it into 11 categories, including Coding, Math, Humanities, STEM, Writing, Summarization, Roleplay, RAG, Multilingual, Reasoning, and QA. Unlike prior benchmarks, which often suffer from low intra-category diversity, the SPEED-Bench Qualitative split explicitly prioritizes semantic diversity. It is designed to evaluate acceptance rates and overall performance of SD algorithms across different task domains.
A Throughput data split: This configuration consists of fixed-length input sequences (1K, 2K, 8K, 16K and 32K) organized into three coarse difficulty categories: low-entropy domains (such as coding), high-entropy domains (such as creative writing), and mixed-entropy domains. The goal is to assess throughput under high concurrency and a wide range of input sequence lengths (ISLs), which are often much longer than the short ISL samples used in many SD benchmarks. As batch size increases, inference often transitions from a compute-bound regime to a memory-bound regime, fundamentally changing the cost-benefit trade-offs of speculative decoding.
| ⚠️ Important: Some examples in the dataset are masked with the sentence "FULL BENCHMARK DATA SHOULD BE FETCHED FROM THE SOURCE USING SPECDEC_BENCH". To access the complete benchmark data, use the accompanying measurement framework, which fetches the full data directly from the sources. |
|---|
Dataset Structure
Each data point contains the following fields:
question_id- unique identifier of the data point in SPEED-Bench.category- data point category- For qualitative can be one of: coding, stem, humanities, math, multilingual, qa, rag, roleplay, reasoning, summarization, writing.
- For throughput can be one of: high_entropy, mixed, low_entropy.
sub_category- a fine-grained category of the data point, if provided by the original data source.turns- a list of prompts, used to form the conversation.source- URL for the original data source.src_id- original ID of the example in the data source.difficulty- the difficuly level of the example, if provided by the original data source.multiturn- a boolean field to indicate whether the data point contain a multi-turn conversation (i.e.len(turns) > 1).
Dataset Creation
Curation Rationale - The Qualitative Split
The goal of the Qualitative split is to measure SD quality, specifically conditional acceptance rates (ARs) and acceptance lengths (ALs), across a wide range of semantic domains.
While extensive evaluation across numerous datasets is theoretically possible, it is tedious, impractical for rapid experimentation, and hinders direct comparisons between different research groups releasing SD algorithms and models. Instead of relying on exhaustive evaluations across disparate datasets, we curate a compact yet highly representative subset designed to maximize semantic diversity. We aggregate data from 18 publicly available sources and organize it into 11 categories, including Coding, Math, Humanities, STEM, Writing, Summarization, Roleplay, RAG, Multilingual, Reasoning, and QA.
Unlike prior benchmarks, which often suffer from low intra-category diversity, the SPEED-Bench Qualitative split explicitly prioritizes semantic diversity.
To achieve this, each candidate prompt is embedded into a dense vector space using a pretrained text embedder (openai/text-embedding-3-small).
We then apply a selection algorithm that minimizes average pairwise cosine similarity within each category.
This ensures that the selected samples span the semantic space as widely as possible, reducing redundancy and increasing evaluation fidelity.
Sampling Algorithm
For each category, our goal is to select a subset of 80 examples that maximizes semantic diversity. The motivation is to produce more reliable acceptance rate measurements for speculative decoding algorithms across varied inputs.
Procedure:
- Embedding computation: We compute embeddings for all examples in the category using the OpenAI
text-embedding-3-smallmodel. - Greedy selection: Starting from a random example, we iteratively select examples that minimize semantic similarity with the already chosen subset, until reaching 80 examples.
- Subset optimization: After the initial selection, we perform swaps between selected and unselected examples to further reduce similarity and enhance semantic diversity.
The table below shows the average semantic similarity of SpecBench examples compared to SPEED-Bench subsets selected using random selection and the greedy selection (as described in the sampling algorithm above).
All values were computed using cosine similarity on embeddings generated with the OpenAI text-embedding-3-small model.
Lower values indicate better semantic diversity. The numbers in parentheses show the relative change compared to SpecBench.
| Category | SpecBench | SPEED (random selection) | SPEED (greedy algorithm) |
|---|---|---|---|
| Math | 0.24 | 0.21 (-12.5%) | 0.15 (-37.5%) |
| Coding | 0.33 | 0.48 (+45%) | 0.16 (-51%) |
| Multilingual | 0.36 | 0.18 (-50%) | 0.06 (-83%) |
| Writing | 0.35 | 0.29 (-17%) | 0.18 (-48%) |
| Humanities | 0.26 | 0.14 (-46%) | 0.12 (-54%) |
| STEM | 0.24 | 0.19 (-21%) | 0.12 (-50%) |
| Reasoning | 0.24 | 0.21 (-12.5%) | 0.14 (-41%) |
| RAG | 0.15 | 0.17 (+13%) | 0.13 (-13%) |
| Roleplay | 0.28 | 0.25 (-11%) | 0.24 (-14%) |
| Summarization | 0.21 | 0.24 (+14%) | 0.11 (-48%) |
| QA | 0.10 | unchanged samples | unchanged samples |
Curation Rationale - The Throughput Split
While the Qualitative split captures draft accuracy, it is insufficient for evaluating system-level speedups. The Throughput split is designed specifically to capture this.
We construct fixed ISL buckets ranging from 1k to 32k tokens, reflecting the growing importance of long-context applications such as coding assistants and retrieval-augmented generation.
For each ISL bucket, prompts are aggregated into three coarse difficulty categories corresponding to low-, mixed-, and high-entropy domains.
To ensure deterministic prefill cost, prompts are either truncated or padded in a controlled manner, while preserving their semantic content.
Each ISL bucket contains 1,536 prompts (512 per difficulty category), providing sufficient volume to construct stable throughput Pareto curves across a wide range of batch sizes.
Importantly, SPEED-Bench avoids the use of random token inputs for throughput benchmarking. Random tokens can severely distort acceptance behavior, expert routing in MoE models, and throughput measurements, leading to overly optimistic conclusions.
Data Sources
We use the following data sources in our benchmark:
| Data Source | Categories | Construction Details |
|---|---|---|
| SpecBench | STEM, Humanities, Writing, Coding, Math, Roleplay, Reasoning, QA, RAG | Used directly from source |
| CNN/Daily Mail | Summarization | Used articles directly from source with instructions to summarize the content. |
| Humanity's Last Exam | STEM, Humanities, Math | Filtered to text-only samples (no images) with exact-match answer type. For STEM: filtered to Physics, CS/AI, Biology/Medicine, Chemistry, Engineering. For Humanities: filtered to Humanities/Social Science category. |
| LiveCodeBench Lite | Coding | Constructed instruction prompts requesting code generation in a randomly selected programming language (Python, Java, C++, Go, JavaScript, Rust). Includes starter code when available. |
| Code Contests | Coding | Constructed instruction prompts requesting program generation in a randomly selected language (Python, Java, C++). Problem descriptions used directly from source. |
| HumanEvalPack | Coding | Used code completion prompts directly from source. |
| RoleBench | Roleplay | Constructed multi-turn roleplay prompts using role descriptions and questions. Questions grouped by role into conversations (1--5 turns). System prompts randomly sampled from 8 prompt templates instructing the model to embody the character. |
| CoSER | Roleplay | Constructed roleplay prompts with character profiles, scenario, and character motivation, only for books that are available in the public domain. |
| WritingBench | Writing | Filtered to English samples. Writing queries used directly as single-turn prompts. |
| Creative Writing V3 | Writing | Expanded prompts by replacing <SEED> placeholders with the seed modifiers provided, creating multiple variations per base prompt. |
| MT-Bench 101 | Reasoning | Filtered to general reasoning and mathematical reasoning tasks. |
| MMLU-Pro | Reasoning | Grouped questions by category and combined multiple questions together to create multi-turn samples. |
| MMATH | Multilingual | Questions used directly from source. |
| OPUS-100 | Multilingual | Constructed translation prompts by prepending "Translate the following text from [source language] to [target language]:". |
| MCIF | Multilingual | Selected prompts for QA, translation, and summarization tasks with long_mixed-prompt format. |
| ChatRAG-Bench | RAG | Constructed prompts with context (concatenated retrieved passages) and multi-turn questions for the hybridial and sqa splits. |
| MCIF | RAG | SUsed English QA prompts with long_mixed-prompt format, grouping questions by document into multi-turn conversations. |
| BAMBOO | Throughput split (writing - high entropy) | Used MeetingPred and ShowsPred subsets. Constructed dialogue completion prompts asking the model to continue conversations. For longer contexts (>16k tokens), concatenated multiple dialogues. Padded/truncated to target token count. |
| Project Gutenberg | Throughput split (writing - high entropy) | Constructed book continuation prompts. Filtered to books with sufficient length and padded/truncated to target token count. |
| WritingBench | Throughput split (writing - high entropy) | Reused English writing prompts from Qualitative Split. Filtered to prompts within 0.7--2x target token count, then padded/truncated. |
| AdaLEval (StackSelect) | Throughput split (NIAH - mixed) | Constructed needle-in-a-haystack prompts asking models to select the most helpful answer from a set of StackOverflow answers and provide explanations for each choice. Padded/truncated to target token count. |
| Humanity's Last Exam | Throughput split (mixed) | Used 50% of HLE data for few-shot prompting. Constructed prompts with category-specific demonstrations sampled from held-out examples, followed by the target question. Padded/truncated to target token count. |
| Long Code Arena | Throughput split (coding - low entropy) | Used project-level code completion subset. Constructed prompts with repository context and file with [COMPLETE] markers for line-level completion. |
| RepoBench Python | Throughput split (coding - low entropy) | Constructed cross-file code completion prompts with repository context snippets and in-file code. Padded/truncated to target token count. |
| RepoBench Java | Throughput split (coding - low entropy) | Same construction as RepoBench Python but for Java code. |
| AdaLEval (TextSort) | Throughput split (sorting - low entropy) | Modified original sorting task to require outputting sorted text segments in order rather than just returning indices. Padded/truncated to target token count. |
Detailed Data Comparison with SpecBench
| Metric | SPEED-Bench | SpecBench |
|---|---|---|
| # Samples per Category | 80 (qualitative), 512 x 3 (throughput) | 10 (for 8 categories), 80 (the rest) |
| # Total Samples | 880 (qualitative), 1536 x 5 (throughput) | 480 |
| # Data Sources | 24 | 5 |
| Avg. Pairwise Similarity | 0.14 | 0.22 |
| # Multiturn Prompts | 167 | 80 |
| Max # Turns | 5 | 2 |
| Subcategories | ✅ | ❌ |
| Difficulties | ✅ (for Math, STEM, Humanities, Coding) | ❌ |
| Long ISLs (16k-32k) | ✅ | ❌ |
| Large batches of fixed-size ISLs | ✅ | ❌ |
| Programming Languages Explicitly Mentioned in Coding | Python (27), CPP (9), Java (10), Go (13), Javascript (11), Rust (3), HTML (1), CSS (1) | Python (3), CPP (1), HTML (1), CSS (1) |
| # Distinct Languages in Multilingual | 23 | 2 |
| Languages in Multilingual | EN, DE, ZH, IT, MG, FR, JA, PT, AR, MK, DA, NL, KO, ES, NN, TH, VI, BN, GU, CS, GD, EU, RU | EN, DE |
| Difficulty level in Math, Humanities and STEM categories | Academic level | High school level |
Examples of similarity matrix between SpecBench and SPEED-Bench for two categories (lower values are better):
Dataset Owner(s):
NVIDIA Corporation
Dataset Creation Date:
February 2026
License/Terms of Use:
GOVERNING TERMS: This dataset is governed by the NVIDIA Evaluation Dataset License Agreement.
ADDITIONAL INFORMATION: MIT for bigcode/humanevalpack, RUCAIBox/MMATH, RUCAIBox/BAMBOO and EQ-Bench. Apache 2.0 for Writing Bench and Spec-Bench. CC BY 4.0 for FBK-MT/MCIF. MIT and Apache 2.0 for tianyang/repobench_python_v1.1, JetBrains-Research/lca-project-level-code-completion and tianyang/repobench_java_v1.1.
NOTICE: The prepare_data.py script which is part of the measurement framework hosted in nvidia/Model-Optimizer GitHub repository automatically fetches and constructs the data from all the source datasets. For each dataset a user elects to use, the user is responsible for checking if the dataset license is fit for the intended purpose.
Intended Usage:
This dataset is designed for LLM engineers and researchers working on evaluating Speculative Decoding (SD) and LLM Inference latency measurements.
Dataset Format
The dataset is distributed in parquet format with a hybrid delivery mechanism to ensure license compliance.
Distribution Structure:
- Part of the samples are directly stored in the parquet files.
- Part of the samples contain a placeholder notice, and a reference to a data source. We provide a dataset building script that fetches data directly from the original host locations. By using the script, the user acknowledges that they are downloading data from third parties and assume responsibility for adhering to their specific license terms.
Dataset Quantification
- Record Count: 880 samples in the qualitative split, 1538 samples in the throughput split.
- Measurement of Total Data Storage ~95MB
Reference(s):
Paper: https://arxiv.org/abs/2604.09557 .
If you found SPEED-Bench helpful, please consider citing the below:
@inproceedings{
abramovich2026speedbench,
title={{SPEED}-Bench: A Unified and Diverse Benchmark for Speculative Decoding},
author={Talor Abramovich and Maor Ashkenazi and Izzy Putterman and Benjamin Chislett and Tiyasa Mitra and Bita Darvish Rouhani and Ran Zilberstein and Yonatan Geifman},
booktitle={Forty-third International Conference on Machine Learning},
year={2026},
url={https://openreview.net/forum?id=Rl2uQlCoQX}
}
Ethical Considerations:
NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our terms of service, developers should work with their internal team to ensure this dataset meets requirements for the relevant industry and use case and addresses unforeseen product misuse.
Please report quality, risk, security vulnerabilities or NVIDIA AI Concerns here.
Dataset Card Contact
Talor Abramovich - talora@nvidia.com
Maor Ashkenazi - mashkenazi@nvidia.com
- Downloads last month
- 78