Skip to content

Dataset · nebius

GitHub SWE tasks for agents

nebius/SWE-rebench-V2

32k tasks from real GitHub issues and PRs across 11 languages for training and evaluating SWE agents.

What's inside

A collection of software engineering text tasks gathered from real GitHub issues and pull requests. Includes 32,079 examples in Python, Go, TypeScript, JavaScript, Rust, Java, PHP, Kotlin, Julia, Elixir, Scala, and other languages. For RL training, a cleaned V2.1 RL version of 13,337 tasks in Harbor format is available.

What you can do with it

  • Train a SWE agent to resolve GitHub issues
  • Perform RL fine-tuning on V2.1 RL
  • Use as a benchmark for LLMs in software engineering
Tasks
Text generation
Languages
English
Size
10K–100K rows
Downloads this month
120.8K
Likes
77

How to load it

from datasets import load_dataset

ds = load_dataset("nebius/SWE-rebench-V2")