Dataset · Lego-X
2699 programming tasks for training RL agents
Lego-X/Lego-RL-2699
Difficulty-filtered executable software development tasks for training and evaluating agentic models via RL.
What's inside
Contains 2,699 English executable tasks filtered by difficulty level. Sourced from GAIR/OpenSWE and presented in two parallel formats for the same records: original OpenSWE strings and ready-to-run Harbor RL environment directories with a training index.
What you can do with it
- Training coding AI agents using RL in ready-to-run executable environments
- Benchmarking and debugging SWE agents on a standardized task set
- Tasks
- Text generation
- Languages
- English
- Size
- 1K–10K rows
- Downloads this month
- 30K
- Likes
- 2
Models trained on it
How to load it
from datasets import load_dataset
ds = load_dataset("Lego-X/Lego-RL-2699")