Mercury: An Efficiency Benchmark for LLM Code Synthesis
Paper • 2402.07844 • Published • 2
data_source stringclasses 1
value | prompt listlengths 1 1 | ability stringclasses 1
value | reward_model dict | extra_info dict |
|---|---|---|---|---|
mercury | [
{
"role": "user",
"content": "You are an expert Python programmer. You will be given a question (problem specification) and will generate a correct Python program that matches the specification and passes all tests.\n\nGiven an m x n matrix, return all elements of the matrix in spiral order.\n\n \nExample 1... | code | {
"style": "rule",
"extraction_method": "mercury_harness",
"ground_truth": "{\"slug_name\": \"spiral-matrix\", \"difficulty\": \"Medium\", \"entry_point\": \"spiralOrder\", \"convert_offline\": \"def convert_offline(case):\\n return case\", \"evaluate_offline\": \"def evaluate_offline(inputs, outputs, expected... | {
"index": "mercury_0",
"split": "eval",
"slug_name": "spiral-matrix",
"difficulty": "Medium",
"n_test_cases": 7,
"n_solutions": 14
} |
mercury | [
{
"role": "user",
"content": "You are an expert Python programmer. You will be given a question (problem specification) and will generate a correct Python program that matches the specification and passes all tests.\n\nYou are given a sorted unique integer array nums.\n\nA range [a,b] is the set of all inte... | code | {
"style": "rule",
"extraction_method": "mercury_harness",
"ground_truth": "{\"slug_name\": \"summary-ranges\", \"difficulty\": \"Easy\", \"entry_point\": \"summaryRanges\", \"convert_offline\": \"def convert_offline(case):\\n return case\", \"evaluate_offline\": \"def evaluate_offline(inputs, outputs, expecte... | {
"index": "mercury_1",
"split": "eval",
"slug_name": "summary-ranges",
"difficulty": "Easy",
"n_test_cases": 7,
"n_solutions": 15
} |
mercury | [
{
"role": "user",
"content": "You are an expert Python programmer. You will be given a question (problem specification) and will generate a correct Python program that matches the specification and passes all tests.\n\nGiven two strings s and t, return the number of distinct subsequences of s which equals t... | code | {
"style": "rule",
"extraction_method": "mercury_harness",
"ground_truth": "{\"slug_name\": \"distinct-subsequences\", \"difficulty\": \"Hard\", \"entry_point\": \"numDistinct\", \"convert_offline\": \"def convert_offline(case):\\n return case\", \"evaluate_offline\": \"def evaluate_offline(inputs, outputs, ex... | {
"index": "mercury_2",
"split": "eval",
"slug_name": "distinct-subsequences",
"difficulty": "Hard",
"n_test_cases": 7,
"n_solutions": 30
} |
mercury | [
{
"role": "user",
"content": "You are an expert Python programmer. You will be given a question (problem specification) and will generate a correct Python program that matches the specification and passes all tests.\n\nGiven an unsorted integer array nums, return the smallest missing positive integer.\n\nYo... | code | {
"style": "rule",
"extraction_method": "mercury_harness",
"ground_truth": "{\"slug_name\": \"first-missing-positive\", \"difficulty\": \"Hard\", \"entry_point\": \"firstMissingPositive\", \"convert_offline\": \"def convert_offline(case):\\n return case\", \"evaluate_offline\": \"def evaluate_offline(inputs, o... | {
"index": "mercury_3",
"split": "eval",
"slug_name": "first-missing-positive",
"difficulty": "Hard",
"n_test_cases": 18,
"n_solutions": 20
} |
mercury | [
{
"role": "user",
"content": "You are an expert Python programmer. You will be given a question (problem specification) and will generate a correct Python program that matches the specification and passes all tests.\n\nThe set [1, 2, 3, ..., n] contains a total of n! unique permutations.\n\nBy listing and l... | code | {
"style": "rule",
"extraction_method": "mercury_harness",
"ground_truth": "{\"slug_name\": \"permutation-sequence\", \"difficulty\": \"Hard\", \"entry_point\": \"getPermutation\", \"convert_offline\": \"def convert_offline(case):\\n return case\", \"evaluate_offline\": \"def evaluate_offline(inputs, outputs, ... | {
"index": "mercury_4",
"split": "eval",
"slug_name": "permutation-sequence",
"difficulty": "Hard",
"n_test_cases": 8,
"n_solutions": 20
} |
mercury | [
{
"role": "user",
"content": "You are an expert Python programmer. You will be given a question (problem specification) and will generate a correct Python program that matches the specification and passes all tests.\n\nGiven a 1-indexed array of integers numbers that is already sorted in non-decreasing orde... | code | {
"style": "rule",
"extraction_method": "mercury_harness",
"ground_truth": "{\"slug_name\": \"two-sum-ii-input-array-is-sorted\", \"difficulty\": \"Medium\", \"entry_point\": \"twoSum\", \"convert_offline\": \"def convert_offline(case):\\n return case\", \"evaluate_offline\": \"def evaluate_offline(inputs, out... | {
"index": "mercury_5",
"split": "eval",
"slug_name": "two-sum-ii-input-array-is-sorted",
"difficulty": "Medium",
"n_test_cases": 8,
"n_solutions": 20
} |
mercury | [
{
"role": "user",
"content": "You are an expert Python programmer. You will be given a question (problem specification) and will generate a correct Python program that matches the specification and passes all tests.\n\nGiven a string num that contains only digits and an integer target, return all possibilit... | code | {
"style": "rule",
"extraction_method": "mercury_harness",
"ground_truth": "{\"slug_name\": \"expression-add-operators\", \"difficulty\": \"Hard\", \"entry_point\": \"addOperators\", \"convert_offline\": \"def convert_offline(case):\\n return case\", \"evaluate_offline\": \"def evaluate_offline(inputs, outputs... | {
"index": "mercury_6",
"split": "eval",
"slug_name": "expression-add-operators",
"difficulty": "Hard",
"n_test_cases": 8,
"n_solutions": 3
} |
mercury | [
{
"role": "user",
"content": "You are an expert Python programmer. You will be given a question (problem specification) and will generate a correct Python program that matches the specification and passes all tests.\n\nYou are given the root of a binary tree containing digits from 0 to 9 only.\n\nEach root-... | code | {
"style": "rule",
"extraction_method": "mercury_harness",
"ground_truth": "{\"slug_name\": \"sum-root-to-leaf-numbers\", \"difficulty\": \"Medium\", \"entry_point\": \"sumNumbers\", \"convert_offline\": \"def convert_offline(case):\\n import lctk\\n inputs, expected = case\\n inputs = lctk.binaryTree(*i... | {
"index": "mercury_7",
"split": "eval",
"slug_name": "sum-root-to-leaf-numbers",
"difficulty": "Medium",
"n_test_cases": 7,
"n_solutions": 16
} |
mercury | [{"role":"user","content":"You are an expert Python programmer. You will be given a question (proble(...TRUNCATED) | code | {"style":"rule","extraction_method":"mercury_harness","ground_truth":"{\"slug_name\": \"license-key-(...TRUNCATED) | {"index":"mercury_8","split":"eval","slug_name":"license-key-formatting","difficulty":"Easy","n_test(...TRUNCATED) |
mercury | [{"role":"user","content":"You are an expert Python programmer. You will be given a question (proble(...TRUNCATED) | code | {"style":"rule","extraction_method":"mercury_harness","ground_truth":"{\"slug_name\": \"gas-station\(...TRUNCATED) | {"index":"mercury_9","split":"eval","slug_name":"gas-station","difficulty":"Medium","n_test_cases":7(...TRUNCATED) |
End of preview. Expand in Data Studio
Mercury (verl efficiency eval set)
The eval split of Elfsong/Mercury
(arXiv 2402.07844; 256 LeetCode-style tasks;
the train split ships no test cases and is not gradable), converted to the verl
rule-reward schema by verl/scripts/data/mercury.py. Source license
CC-BY-NC-4.0 (non-commercial) -- this conversion keeps that license.
Every row's ground truth carries the full official scoring contract: entry
point, the task's convert_offline/evaluate_offline hooks (lctk linked-list /
binary-tree conversions included), the materialised test cases, and the
reference solution pool. All 256 rows verified at build time: at least one
reference solution passes the grading harness locally.
Scoring (verl/verl/utils/reward_score/mercury.py): the extracted
class Solution runs all test cases through the task's own hooks in ONE timed
firejail execution, reproducing the official sandbox
(github.com/Elfsong/Mercury src/sandbox.py). On a full pass, the official
Beyond percentile is computed against the task's reference solutions
re-executed on the same machine (cached per task);
reward = 0.5*pass + 0.5*beyond. Eval recipe: verl/recipe/run_mercury_eval.sh.
Known compromises vs the official evaluator
- At most
MERCURY_MAX_REFS(default 8) of the ~18 reference solutions are measured per task, sampled evenly across the LeetCode-runtime ordering, to bound reward-time cost. Beyond = clip((max_rt - rt)/(max_rt - min_rt), 0, 1)-- the paper's formula. The reference implementation clamps the numerator at 1 second before dividing (treated here as a bug).- Runtimes come from a single un-repeated pass (official parity); on this benchmark's small generated inputs, sub-millisecond loop times make Beyond noisy at the per-task level. Aggregate Beyond@1 over 256 tasks is stable.
- Execution runs under firejail rlimits, not the official in-process reliability guard.
- Downloads last month
- 23