---
task_categories:
  - text-generation
  - text2text-generation
language:
  - en
tags:
  - coding
  - code generation
  - atcoder
  - programming contests
pretty_name: Atcoder's contests
size_categories:
  - 100K<n<1M
---
## Dataset Description

- **Homepage:** None
- **Repository:** <https://huggingface.co/datasets/Nan-Do/atcoder_contests>
- **Paper:** None
- **Leaderboard:** None
- **Point of Contact:** [@Nan-Do](https://github.com/Nan-Do)

### 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](mailto:tarin.f@gmail.com)**)

### 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](https://atcoder.jp/)

### 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.