Dataset Summary
This dataset aims to facilitate the creation of sophisticated, multi-turn dialogue datasets focused on coding for Large Language Models (LLMs).
It also serves as a robust foundation for problem-solving in Large Language Models (LLMs).
The dataset includes both accepted and failed solutions from Atcoders's (ABC) contests.
In total, it features 1911 unique problems and 384,536 submissions across over 50 different programming languages.
It covers contests from ABC contest 50 (previous contests only have the statements in japanese) up to ABC contest 350 (April 20, 2024).
(If you are interested in accessing the dataset please contact me here)
Multi-turn dialogue
A multi-turn dialogue can be structured as follows:
- [USER] - Given the problem statement below, please provide a solution in {lang}.
- [SYSTEM] - Sure! Here's a solution in {lang}. [...] (Using a TLE submission)
- [USER] - The provided solution is too slow. Can you suggest a faster implementation?
- [SYSTEM] - Of course! Here's an optimized solution. [...] (Using an ACC submission)
This template is versatile and can be adjusted to accommodate various scenarios and/or expanded to facilitate high-quality conversations for training Large Language Models (LLMs).
Languages
The problem descriptions are in English. The dataset includes submissions in various programming languages, each with specific language and compiler versions. For example, for C++:
| Version | Compiler |
|---|---|
| C++ 20 | Gcc 12.2 |
| C++ 17 | Gcc 12.2 |
| C++ 23 | Gcc 12.2 |
| C++ 23 | Clang 16.0.6 |
| C++ 20 | Clang 16.0.6 |
| C++ 17 | Clang 16.0.6 |
| C++ | GCC 9.2.1 |
Here is a summary of the languages used for the submissions:
| Programming Language | Total |
|---|---|
| cpp | 236325 |
| python3 | 106360 |
| rust | 11134 |
| java | 6652 |
| csharp | 4141 |
| c | 3867 |
| golang | 2970 |
| javascript | 1478 |
| php | 863 |
| kotlin | 337 |
| typescript | 330 |
| swift | 310 |
| common lisp | 176 |
| scala | 117 |
| dart | 36 |
| vim | 19 |
Status
The status indicates the outcome of each submission. Below is an explanation of the status values, their meanings, and their totals:
| Value | Meaning | Total |
|---|---|---|
| AC | Submission Accepted | 338811 |
| WA | Wrong Answer | 34964 |
| RE | Runtime Error | 5858 |
| TLE | Time Limit Exceeded | 4861 |
| MLE | Memory Limit Exceeded | 42 |
Data Splits
There are no splits (Only training).
Dataset Creation
Jun of 2024
Curation Rationale
This dataset was designed to enhance the coding and problem-solving capabilities of LLMs by providing sophisticated multi-turn dialogues focused on coding.
Source Data
The source of the dataset is Atcoder's contests
Annotations
The dataset includes the following columns:
- contest_id: ID of the contest (e.g., abc-250).
- problem_statement: Markdown explanation of the problem.
- task: Task identifier within the contest (e.g., A).
- lang: Programming language used for the solution (see previous table for details).
- score: Score awarded for the task (e.g., 100, or 0 if failed).
- code_size: Size of the code in bytes.
- status: Outcome of the submission (see previous table for details).
- execution_time: Time taken by the submission in milliseconds.
- execution_memory: Memory used by the submission in kilobytes.
- code: Code that solves the problem.
- Downloads last month
- 40