Dataset · FineEnvs
Tabular tasks for small model RL
FineEnvs/SmolDataEnvs
5.5K data analysis tasks: table, question, and ground truth answer.
What's inside
The dataset contains over 5.5K data analysis and code tasks in a plug-and-play format: each row includes a real tabular dataset, a question about it, and a gold answer. The authors demonstrate training a 2B model, 144 held-out tasks, and two runs of 5,000 tasks — shuffled and curriculum-ordered from simple to complex. Modalities are tabular and text; metadata size is 1K–10K.
What you can do with it
- RL training for small LLMs to solve tabular tasks
- Testing curriculum learning and sensitivity to task order
- Evaluating table QA and data analysis without an external runtime
- Tasks
- Question answering, Table question answering
- Size
- 1K–10K rows
- Downloads this month
- 89.6K
- Likes
- 106
Models trained on it
How to load it
from datasets import load_dataset
ds = load_dataset("FineEnvs/SmolDataEnvs")