LiveCodeBench (Jain 2024年)(ライブコードベンチ)
2024年Naman Jain et al. (UC Berkeley)発表LiveCodeBench・Industry-leading live coding contamination-free benchmark + Industry-leading LeetCode+CodeForces+AtCoder + Industry-leading rolling update + Industry-leading contamination-free coding benchmark standard。
概要
LiveCodeBench は、2024年Naman Jain et al. (UC Berkeley)発表LiveCodeBench・Industry-leading live coding contamination-free benchmark 2024年 + Industry-leading contamination-free coding benchmark standard position確立。LiveCodeBench specifications = Industry-leading live coding contamination-free benchmark (Industry-leading LiveCodeBench Jain 2024 + Industry-leading live coding + Industry-leading UC Berkeley flagship) + Industry-leading LeetCode+CodeForces+AtCoder + Industry-leading rolling update + Industry-leading contamination-free coding benchmark standard。
主な特徴・仕組み
- Authors: Industry-leading Naman Jain + King Han + Alex Gu + Wen-Ding Li + Fanjia Yan + Tianjun Zhang + Sida Wang + Armando Solar-Lezama + Koushik Sen + Ion Stoica (UC Berkeley + MIT + Cornell)
- Benchmark: Industry-leading LiveCodeBench live coding contamination-free 2024
- Year: 2024年 (arXiv 2024年3月 + rolling update 2024年-)
- Paper: Industry-leading "LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code" arXiv 2403.07974
- Method Type: Industry-leading live coding contamination-free benchmark
- Live Coding: Industry-leading live coding signature
- 3 Sources: Industry-leading LeetCode + CodeForces + AtCoder competitive programming sources
- Rolling Update: Industry-leading rolling update by date signature (new problems monthly)
- Contamination-Free: Industry-leading contamination-free training data leak avoidance
- 4 Scenarios: Industry-leading 4 scenarios (Code Generation/Self-Repair/Code Execution/Test Output)
- Industry Adoption: Industry-leading Claude/GPT/Gemini/DeepSeek coding benchmark adoption
- UC Berkeley: Industry-leading UC Berkeley Ion Stoica + Koushik Sen authors
- Industry-Leading: Industry-leading UC Berkeley LiveCodeBench contamination-free coding 2024
スペック比較表
| LLM Evaluation Benchmark (2023-2024) | Year | Org | Type | Industry Position |
|---|---|---|---|---|
| LiveCodeBench | 2024 | Jain et al. UC Berkeley | Live coding contamination-free | Industry-leading contamination-free coding |
| SWE-bench Verified | 2024 | OpenAI | Real-world coding agent | Industry-leading coding agent standard |
| GPQA Diamond | 2023 | Rein et al. NYU | PhD-level Q&A | Industry-leading PhD-level reasoning |
| MMLU-Pro | 2024 | TIGER-Lab Wang | Enhanced MMLU | Industry-leading MMLU-Pro reasoning |
| AIME 2024 | 2024 | AMC math competition | Math reasoning competition | Industry-leading math reasoning |
具体例・対応製品
- LiveCodeBench (2024年, Jain et al. UC Berkeley): Industry-leading UC Berkeley contamination-free coding
- Industry-leading live coding signature: Industry-leading live coding
- Industry-leading LeetCode + CodeForces + AtCoder 3 sources: Industry-leading 3 competitive programming sources
- Industry-leading rolling update by date signature: Industry-leading rolling update by date
- Industry-leading 4 scenarios (Code Generation/Self-Repair/Code Execution/Test Output): Industry-leading 4 scenarios
- 競合 SWE-bench Verified + GPQA Diamond + MMLU-Pro + AIME 2024: Industry-leading benchmark competitors
選び方・注意点
LiveCodeBench は「Industry-leading live coding + LeetCode+CodeForces+AtCoder」「Industry-leading rolling update + contamination-free + UC Berkeley」用途のIndustry-leading 2024年 UC Berkeley product。Industry-leading live coding (Industry-leading live coding signature + Industry-leading LiveCodeBench live signature + Industry-leading time-dated problem selection) で Industry-leading live coding + Industry-leading LiveCodeBench + Industry-leading time-dated problem。Industry-leading 3 sources (Industry-leading LeetCode + CodeForces + AtCoder competitive programming sources + Industry-leading LiveCodeBench 3 sources signature + Industry-leading multi-platform coverage) で Industry-leading 3 sources + Industry-leading LiveCodeBench + Industry-leading multi-platform coverage。Industry-leading rolling update (Industry-leading rolling update by date signature new problems monthly + Industry-leading LiveCodeBench rolling signature + Industry-leading contamination-free design) で Industry-leading rolling update + Industry-leading LiveCodeBench + Industry-leading contamination-free design。Industry-leading 4 scenarios (Industry-leading 4 scenarios Code Generation/Self-Repair/Code Execution/Test Output + Industry-leading LiveCodeBench 4-scenario signature + Industry-leading comprehensive coding evaluation) で Industry-leading 4 scenarios + Industry-leading LiveCodeBench + Industry-leading comprehensive coding。Industry-leading UC Berkeley (Industry-leading Naman Jain + King Han + Alex Gu + Wen-Ding Li + Sida Wang + Armando Solar-Lezama + Koushik Sen + Ion Stoica UC Berkeley + MIT + Cornell + Industry-leading arXiv 2403.07974 + Industry-leading 2024年3月 release + rolling update 2024年- + Industry-leading widespread Claude/GPT/Gemini/DeepSeek coding benchmark) で Industry-leading UC Berkeley + Industry-leading Jain+Han+Gu+Solar-Lezama+Stoica + Industry-leading 2403.07974 + Industry-leading 2024年3月 + Industry-leading widespread Claude/GPT/Gemini/DeepSeek。但しIndustry-leading SWE-bench Verified + GPQA Diamond + MMLU-Pro + AIME 2024 competition で Industry-leading 4-benchmark competitors vs LiveCodeBench + Industry-leading live coding + LeetCode+CodeForces+AtCoder 3 sources + rolling update by date + contamination-free training data leak avoidance + 4 scenarios Code Generation/Self-Repair/Code Execution/Test Output + UC Berkeley + MIT + Cornell + Jain+Han+Gu+Solar-Lezama+Stoica + arXiv 2403.07974 2024年3月 + rolling update 2024年- unique advantage adoption alignment必須。
関連用語との違い
- vs SWE-bench Verified (OpenAI 2024): LiveCodeBenchはcompetitive programming live coding contamination-free・SWE-bench Verifiedはreal-world GitHub issue agent
- vs HumanEval (Chen 2021): LiveCodeBenchはlive rolling contamination-free competitive・HumanEvalは164 static Python functions
- vs GPQA Diamond/MMLU-Pro (2023-2024): LiveCodeBenchはcompetitive coding contamination-free・GPQA/MMLU-Proはアカデミック reasoning
よくある質問(FAQ)
Q1: LiveCodeBench vs SWE-bench Verified 違いは? A: LiveCodeBench (Industry-leading live coding + LeetCode+CodeForces+AtCoder 3 sources + rolling update by date + contamination-free + 4 scenarios Code Generation/Self-Repair/Code Execution/Test Output + UC Berkeley+MIT+Cornell + Jain+Han+Gu+Solar-Lezama+Stoica + arXiv 2403.07974 + 2024年3月 + rolling update) vs SWE-bench Verified (Industry-leading SWE-bench human-verified subset + 500 GitHub issue tasks + 12 Python repositories Django/Sympy/etc + real-world coding + Princeton 2310.06770 base + OpenAI 2024年8月)・Industry-leading live coding + LeetCode+CodeForces+AtCoder + rolling update + contamination-free + 4 scenarios + UC Berkeley + 2024年3月 = LiveCodeBench + Industry-leading human-verified + 500 GitHub issues + 12 Python repos + real-world + Princeton + OpenAI + 2024年8月 = SWE-bench Verified preference judgment (competitive programming live vs real-world GitHub differ)。
Q2: Industry-leading live coding + rolling update value は? A: Industry-leading live coding + rolling update (Industry-leading live coding signature + Industry-leading rolling update by date signature new problems monthly + Industry-leading LiveCodeBench dual signature)。
Q3: Industry-leading 3 sources + 4 scenarios + UC Berkeley value は? A: Industry-leading 3 sources + 4 scenarios + UC Berkeley (Industry-leading LeetCode + CodeForces + AtCoder 3 sources + Industry-leading 4 scenarios Code Generation/Self-Repair/Code Execution/Test Output + Industry-leading UC Berkeley + MIT + Cornell Jain+Han+Gu+Solar-Lezama+Stoica)。
まとめ
LiveCodeBench = 2024年Jain et al. UC Berkeley + MIT + Cornell発表のlive coding contamination-free benchmark。Industry-leading live coding signature + Industry-leading LeetCode + CodeForces + AtCoder competitive programming 3 sources + Industry-leading rolling update by date signature new problems monthly + Industry-leading contamination-free training data leak avoidance + Industry-leading 4 scenarios Code Generation/Self-Repair/Code Execution/Test Output + Industry-leading widespread Claude/GPT/Gemini/DeepSeek coding benchmark + Industry-leading Naman Jain + King Han + Alex Gu + Wen-Ding Li + Sida Wang + Armando Solar-Lezama + Koushik Sen + Ion Stoica UC Berkeley + MIT + Cornell + Industry-leading "LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code" arXiv 2403.07974 + Industry-leading 2024年3月 release + rolling update 2024年- + Industry-leading UC Berkeley LiveCodeBench contamination-free coding 2024 position確立。